Lombardi_1990_The_nonlinear_organization_of_the_affricate

LINDA LOMBARDI
THE NONLINEAR ORGANIZATION OF

THE AFFRICATE*
This paper proposes that affricates are composed of [-cont] and [+ cont] specifications
which are unordered at underlying representation and throughout the phonological
derivation, although they are ordered phonetically. Previous autosegmental treat-
ments of the affricate, which include this ordering as part of underlying representation,
can handle edge effects, which are cases where rules refer to only the adjacent value
of [con@ But the ordered representation cannot account for anti-edge effects: those
rules which involve the nonadjacent value of [cont]. Moreover, it predicts that ordering
of the gestures is a possible phonological contrast, but this is incorrect, as no language
has [+cont][-cont] single segments. As predicted by the unordered representation,
all apparent cases of phonological edge effects can simply be stated as rules referring
to a value of [cont]; all rules which show true edge effects occur after the values are
phonetically ordered.
O. INTRODUCTION
Affricates are segments which are composed, on the surface, of ordered

[-cont][+cont] gestures. Previous autosegmental analyses have included
this ordering as part of the underlying representation. For example, Sagey
(1986) represents the affricate as a single segment with two ordered values
of the feature [cont]:
(1)
X
I • Root
/ix
[-cont] [+cont]
This representation is proposed in order to account for rules that show

'edge effects'. These are rules in which affricates behave as stops with
regard to rules sensitive to their left edges, and as fricatives with regard
* I would like to express my gratitude to John McCarthy, to whom I and this paper owe a
not inconsiderable debt. Thanks also to Francois Dell, F. Roger Higgins, Jos6 Hualde,
Michael Kenstowicz, Juliette Levin, Lisa Selkirk, Alison Taub, and three anonymous re-
viewers for helpful suggestions at various stages. This work was supported in part by SDF
grant 626 awarded to John McCarthy.
Natural Language and Linguistic Theory 8: 375425, 1990.

© 1990 Kluwer Academic Publishers. Printed in the Netherlands.
376 LINDA LOMBARDI
to rules sensitive to their right edges. Sagey (1986) gives the English plural
rule as an example of an edge effect. Assume that the English plural rule
inserts schwa between the plural ending and a word-final strident [+ cont]. ~
The plural ending, and thus the context of the rule, is on the right edge
of the affricate. This is the [+cont] edge of the affricate on the surface,
and the rule inserts schwa after an affricate (churches) as well as after a
fricative (bushes).
A representation in which there are two ordered features is referred to
as a contour seg/nent. Sagey distinguishes this from a complex segment,
such as [k'-0], which has two places of articulation that have no underlying
ordering.
If the values of [cont] in the affricate are ordered, it follows that phono-
logical processes will show edge effects. However, there are also many
processes applying to affricates that show the opposite of edge effects: for
example, a rule that applies after stops applies after affricates as well,
although the [-cont] value of the affricate is presumed to be on the left
(Archangeli and Pulleyblank 1987). This is a difficulty for the contour
segment analysis.
In addition, a representation where the values of [cont] are ordered
incorrectly predicts a nonexistent type of consonant. Lexicat entries should
not contain predictable specifications (Kiparsky 1982); Prince (1987) and
McCarthy (1989) have extended this idea to linear order relations among
segments or features. If linear order is stipulated underlyingly, then this
predicts that the order should be contrastive. For example, in languages
that have contour tones, the values of H and L must be ordered, because
the contour tones H L and LH are distinct. This is not the case with
affricates; the ordering of the stop and fricative parts is never contrastive,
so that [g~] is not a possible affricate (as was pointed out by Kaye (1985).
There are no single [+cont][-cont] segments.
This paper will demonstrate that the values of the feature [continuant]
must be unordered in the underlying representation of the affricate? I
will argue for the representation in (2), in which the two values of [cont]
are unordered and on separate tiers. The details of this representation
will be made more precise in section 1 below. Aside from the feature
1 I assume this analysis here and in section 2 for expository purposes only; see section 4.3
for a detailed discussion of the English plural rule.
2 Sagey also considers prenasalized stops to be contour segments - segments with two
ordered values of [nasal]. The processes that prenasalized stops participate in are not all
directly comparable to the processes involving affricates. While the issues considered here
should also be considered for prenasalized stops, this paper does n o t make any predictions
about their representation. See Mester (1986, p. 45) for an argument that prenasalized stops
must be unordered.
THE NONLINEAR ORGANIZATION OF THE AFFRICATE 377
[cont], the remainder of the feature geometry assumed here is that of

Sagey (1986) with the modifications of McCarthy (1988). In general the
details of the rest of the geometry have no effect on the arguments
presented herein, except for one: the lack of a Supralaryngeal node, which
will be discussed in section 5.
(2) R ° ° t [ c°ns]son
j
Laryngeal node
[-nasal]
[+cont] [-cont]
Place
With this underlying representation, as will be shown in more detail in
sections 2 and 3, rules involving affricates that do not show edge effects
will be quite simple to analyze. They will be stated as requiring a particular
value of [cont] in the structural description; since the value will be present
in .the affricate, the rule will apply to it. Since the values of [cont] are not
ordered, both values are adjacent to the contexts on either side of the
affricate. In addition, there are phenomena that can only be analyzed if
the values are unordered, as will be shown in sections 3.4 and 3.5.
Although the values of [cont] are unordered at the level of underlying
representation, they are ordered at tho phonetic surface, since the stop
and fricative gestures are performed in that order. The phonetic rule that
orders the values is universal, since no language has backwards affricates.
Since the values are ordered phonetically, phonetic processes are still
predicted to show edge effects. 3 However, as I will show, phonological
processes do not show edge effects, as predicted by this representation.
1. FURTHER DETAILS OF THE FEATURE [CONTINUANT]
Current theories of feature geometry (Mester 1986, Sagey 1986, McCarthy

1988) represent many features as privative: only one value of the feature
3 It should be noted that this paper will mainly deal with the behavior of affricates in
languages where the affricate is contrastive with both stops and fricatives. In many languages
an affricate is found in places of articulation where you would expect a stop in the consonant
system. For example, English does not have an alveopalatal stop, but only an alveopalatal
affricate. In such languages, it is possible that the affricate is underlyingly underspecified for
[+cont], since it does not need the [+cont] feature to make it distinctive from any other
[-cont] segment at that place of articulation; the phonology of the language must be
examined to see if the affricate patterns with fricatives and must be specified [+cont]. Section
4.3 gives some evidence that English affricates are specified [+cont].
378 LINDA LOMBARDI
is present, and segments that do not have that value lack the property
designated by that feature. For instance, a coronal segment is specified
[cor] and a labial segment is [lab], but it is impossible to describe coronals
as [-lab]; rather, they are not specified for any value of [lab]. Considering
a feature to be privative is a separate issue from assuming a theory of
radical underspecification, where only one value of a feature is present in
underlying representation and the opposite value is filled in later in the
derivation. If a feature is privative, the negative value does not exist at
underlying representation or at surface representation: a [cor] segment
never receives a [-lab] specification.
Affricates undergo processes that apply to stops and processes that
apply to fricatives; they must have both the value of [cont] that stops
have, and the value that fricatives have, yet be distinct from both. Thus
we cannot allow [cont] to be a single-valued feature, with fricatives speci-
fied [cont] and stops lacking a specification for [cont]. Combining [cont]
and 'unspecified for [cont]' would yield a segment that was just specified
[cont], which is the representation of a fricative and cannot also be the
representation of an affricate.
So it is necessary to retain two values of [cont] in underlying representa-
tions. However, it is possible for the plus and minus value of this feature
to cooccur, in the case of affricates. The implication of an equipollent
feature is that the two values are opposed, arid cannot coexist. Other
features that are considered to be equipollent - [anterior], for instance -
do not allow both values to occur on a single consonant. But the values
of [cont] can coexist; this is not consistent with the behavior of other
equipollent consonant features. 4
However, single-valued features, such as Place features, can coexist in
complex segments (Sagey 1986). There is no ordering between the Place
features in such segments, since all features are on separate tiers, and
there is no temporal ordering between features on separate tiers in a single
segment. The two values of [cont] in an affricate are, I claim, like separate
Place features, unordered and represented on separate tiers. That is, the
two 'values' of [cont] behave exactly as would be predicted by the hypoth-
esis that they are actually two independent single-valued features. These
features could be called [stop] and [cont], corresponding to [-cont] and
[+cont] respectively. 5
In this view, then, affricates are a type of complex segment, with respect
4 Opposite values of equipollent vowel features can occur in diphthongs; these present a
number of special problems of their own (see Hayes 1988; Kaye 1985).
s This is not a new proposal. Ladefoged (1971) proposes two privative features [stop] and
[fricative] also, although for different reasons.
to manner rather than place. The features [stop] and [cont] are unordered
and on separate tiers in the same way that, say, [labial] and [coronal] are.
This representation will make the correct predictions about what types
of segments exist: segments that are [stop], those that are [cont], and
those that are [cont] and [stop]. This is identical to the prediction made
by single valued Place features; for example, there are segments that are
[labial], segments that are [dorsal], and segments that are [lab] and [dor]
(labiovelars). Segments that are not specified for either [stop] or [cont]
are also a logical possibility. I will assume that these will be segments with
no Place features, since it is impossible to have articulation at a particular
place unless the manner of articulation is specified. 6
If [stop] and [cont] are separate features, the possibility that they are
equipollent features must also be considered. The possible combinations
of equipollent [stop] and [cont] are given in (3):
(3)a. [-stop, +cont] fricatives
b. [+stop, -cont] stops
c. [+stop, +cont] affricates
d. [-stop,-cont] ?
Equipollent features predict an unknown type of segment that has the
features specifications in (3d). This type of segment would pattern with
stops and with fricatives: all rules that apply to [-cont] would affect it,
and so would all rules applying to [-stop]. I know of no evidence that
such a segment exists; thus, the more restrictive version of the theory,
single-valued (i.e., privative) [stop] and [cont], is to be preferred.
The picture in (2) shows the two values of [cont] attached separately
under the Root node. It is possible that the features [stop] and [cont] are
sisters under an abstract Class node (Clements 1985), possibly a Stricture
node. The existence of an abstract node predicts that there would be
spreading and delinking of the entire Stricture node, which would predict
spreading of affricatehood. While the latter is not attested, this may be a
result of the fact that spreading of values of [cont] occurs only in rather
special circumstances (see Mascar6 (1984), and Rosenthall (1988) for an
analysis of postnasal hardening as spreading of [-cont]). (For the same
reason, it is difficult to test the prediction that [stop] and [cont] would
each be transparent to the spreading of the other.) The presence of a
Stricture node might help to account for the fact that [stop] and [cont] do
~ Selkirk (1988) proposes that the Place features are dependants of the feature [cont]. If
this is correct, the assumption that segments with no value of [cont] will be segments without
Place features would follow.
380 LINDA LOMBARDI
seem to have a special relationship to one another; for example, weaken-

ing and strengthening processes often involve [stop] being replaced by
[cont] and vice versa. I will leave this question open, since the presence
or absence of an abstract node has no significance for the main argument
of this paper.
For the sake of familiarity, I will continue to use the terms [+/-cont].
However, it should be clear from these arguments that these must actually
be considered to be names for two features on separate tiers. Strictly,
these features should be called [stop] and [cont] throughout the remainder
of this paper, but I will refer to the features using the familiar terms so
that the arguments about ordering can be followed easily.
2. PHONOLOGICAL PROCESSES INVOLVING AFFRICATES
Two types of phonological rules involving affricates have been distin-

guished: rules showing edge effects, and rules showing anti-edge effects.
Rules showing edge effects are rules that appear to be sensitive to whether
the context for the rule is on the [-cont] or [+cont] edge of the affricate.
Thus this includes two types of rules:
1. Rules that have their context on the right edge of the affricate, which
is the [+cont] edge, and affect [+cont] segments. These rules apply to
fricatives and affricates.
2. Rules that have their context on the left edge of the affricate, which
is the [-cont] edge, and affect [-cont] segments. These rules apply to
stops and affricates.
("Affect" and "apply to" can be replaced by "conditioned by", for rules
that have affricates in their context.)
Rules showing anti-edge effects are rules that are not sensitive to the
edge of the affricate that the context is on. These are rules that see the
value of [cont] that is n o t adjacent to the context of the .rule (under the
assumption that the values are ordered). S o this includes two types of
rules, opposite to the edge effect rules.
1. Rules that have their context on the right edge of the affricate, but
that are rules that affect [-cont] segments, and apply to both stops and
affricates.
2. Rules that have their context on the left edge of the affricate, but
that are rules that affect [+cont] segments, and apply to both fricatives
and affricates.
These rules are a problem for the contour segment analysis. However,
processes that show anti-edge effects are quite simple to state if the values
of [cont] in the affricate are underlyingly unordered. Moreover, processes
that show apparent edge effects can still be analyzed as well. Consider
the diagram in (4), which shows an affricate, with the values of [cont] on
separate tiers, and the contexts before and after the affricate, represented
by A and B.
(4) [+cont]
I
A • Root B
I
[-cont]
If a rule has the context A, the context is adjacent to the surface [-cont]
edge; but there are anti-edge-effect rules that apply to [+cont] segments,
have their context as in A in (4), and yet apply to affricates. If we assume
the values of [cont] are unordered, a rule with the context A that applies
to [+cont] segments will necessarily apply to affricates, since in this repre-
sentation, context A is adjacent to a [+cont] segment. Likewise, a rule
with context B that applies to [-cont] segments will also apply to affri-
cates. Thus, anti-edge effects with affricates are simply normal phonolog-
ical rules acting on a segment with a particular feature specification. No
special machinery is needed.
Some apparent cases of edge effects are also subject to the same analy-
sis. A rule that has its context on the right side of the affricate and that
applies to [+cont] segments may appear to be sensitive to the ordering of
the values of [cont]. But like the anti-edge effects, such rules can simply
be stated as applying to a segment with a particular value of [cont], which
the affricate will have; ordering is not necessary to the analysis. In (4), a
rule with context A (on the left side of the affricate) can also apply to
[-cont] segments, and a rule with context B (on the right side) can apply
to [+cont] segments.
Consider again the English plural rule as discussed in the Introduction.
This rule appears to be sensitive to the right edge of the affricate, since
the context is on the right edge of the affricate, and it treats affricates
and fricatives identically, inserting schwa after both (churches, bushes).
But if the rule inserting schwa is simply stated as applying to a [+cont]
segment, it will apply to affricates and fricatives, which are both [+cont],
and it will not apply to stops, which are not [+cont]. Thus, in many cases
the difference between edge effects and anti-edge effects is illusory.
382 LINDA LOMBARDI
Given the representation of affricates in (2), rules that operate on

underlying representations should never show true edge effects. Phonolog-
ical rules that apply to [-cont] segments should always apply to both stops
and affricates; phonological rules that apply to [+cont] segments should
always apply to both fricatives and affricates, regardless of which side the
context for the rule is on. Only phonetic rules, which operate on the
ordered surface representation of the affricate, will show true edge effects.
3. E V I D E N C E FOR THE T H E O R Y
3.1. Yucatec and Basque
Basque, as discussed by Archangeli and Pulleyblank (1987) and Hualde
(1987), has a rule that deletes stops that immediately precede another
stop. (Their analyses of this process will be discussed in a later section.)
This rule also applies to affricates, in which case the effect is that the
affricate becomes a fricative. This is an example of a rule involving affri-
cates that does not show an edge effect. Although the context for the rule
is on the right-hand side of the affricate - that is, the fricative edge -
nevertheless the rule applies to affricates.
(5) Basque
(from Hualde 1987, Archangeli and Pulleybank 1987)
z = [s], s = [g], x = [g], tz = [c], ts = [6], tx = [~]
a. Stop + stop:
/bait naiz/ bai naiz 'since I am'
/oroit + men/ oroimen 'remembrance'
/guk pitzu/ gu piztu 'we light'
/ardiek nituen/ ardie nituen 'we had sheep'
/bat paratu/ ba paratu 'put one'
/bat traban/ ba traban 'one stuck'
/bat kurri/ ba kurri 'run one'
b. Affricate + stop:
/hitz + tegi/ hiztegi 'dictionary'
/hitz + keta/ hizketa 'conversation'
/haritz + mendi/ harizmendi 'oak mountain'
/haritz + ki/ harizki 'oak wood'
/hotz + tu/ hoztu 'a cold'
A similar process occurs in Yucatec Maya (Straight (1976)): Stops be-
come [h] before homorganic stops an.d affricates; affricates become frica-
tives before homorganic stops and affricates. In other words, stops appear
to lose everything but their laryngeal node, and affricates lose their entire
stop portion, before a stop or an affricate. (Straight unfortunately does
not give an example of affricate before affricate, but his description is
quite clear.)
(6) YucatecMaja
a. taa0 k pak'ik k kool ~ taarj k pak'ik h kool 'we're planting our
clearing'
b. le9 iI3 w ot go ~ le~ i0 w oh go 'that house of mine/my house
there'
c. tun kolik k'aag---~ tun kolih k'aag 'he's clearing bush'
d. ~uc t i0 w i 6 ~ ~us t i0 w i6 'I like it' (lit., 'goodness is at my
eye.')
e. c'u ho~o~ t i k ~ c'u ho~o~ tik 'he scratched it'
("Homorganic" in this rule refers only to the major articulator features -
two coronal consonants are homorganic regardless of whether one of them
is also anterior; so presumably the deletion rule is preceded by a rule of
assimilation. There happen to be no noncoronal affricates in the langu-
age .)
The Basque and Yucatec rule can be formalized as the deletion of
the feature [~cont] before another [-cont], presumably an effect of the
Obligatory Contour Principle (OCP) (Leben 1973; Goldsmith 1976; Mes-
ter 1986; McCarthy 1986). In Yucatec, as in (7), it will apply when the
articulator nodes are identical.
(7)
[-cont] Place [-cont]
The rule will apply to both stops and affricates in both context and focus
because it is a rule that applies to a segment with a [-cont] specification.
Because there is no ordering of the feature values, the feature [-cont] in
the affricate will be adjacent to the rule context, despite the fact that the
context is on the right-hand side of the affricate.
The rule Delete [-cont] in the appropriate context will turn affricates
into fricatives, since an affricate from which only'[-cont] is deleted has
all of the features of the corresponding fricative, and is a well-formed
segment of the language. Then, what effect does this rule have on stops?
When stops lose their [-cont] feature, the segments will have place Of
articulation features, but no specification for manner of articulation. This
384 LINDA LOMBARDI
will result in an ill-formed segment. As explained above in section 1,

segments with no specification for a value of [cont] will be segments with
no Place features; it is impossible to have articulation at a certain Place
if there is no specification for the manner in which the articulation is
carried out. Thus, when the stop loses its manner feature, it will delete,
since the remaining features do not constitute a possible segment of the
language.
The Mayan case shows that in fact the effect of this rule is actually
deletion of the Place features, and that deletion of the entire segment in
Basque is a result of independent factors. Note the different result the
rule has in Yucatec and Basque:
Basque: stops delete, affricates ~ fricatives
Yucatec: stops---> [h], affricates--> fricatives
If [-cont] is deleted from a stop, there is no specification for type of
closure, and so it will be articulatorily impossible to realize the Place
features. However, the remaining laryngeal features are the features of
[h]. 7 This [h] will appear on the surface in Yucatec, where syllable-final
[h] is possible. In a without syllable-final [h], like Basque, or a language
without [h], the result will be deletion of the stop due to the ill-formedness
of the segment or the syllable structure. Thus, in both languages, deletion
of [-cont] causes deletion of Place; in Basque, complete deletion of the
stop is a side effect due to other characteristics of the language.
3.2. Turkish
In Turkish, (Clements and Keyser 1983, Archangeli and Pulleyblank 1987,
Lewis 1967, Nemeth 1962), stops and affricates, but not fricatives, are
devoiced in syllable-final position.
(8) Turkish
NOM
sebep sebepler (PL) sebebi (POSS) 'reason'
pabu6 pabu~lar (PL) pabu)'u (POSS) 'slipper'
deniz denizler (PL) denizi (POSS) 'see'
sepet sepedi (ACC) 'basket'
kitap kitaba (DAT) 'book'
7 See Steriade (1987), Clements (1985), Lass (1976), McCarthy (1988) for various discussions
of reduction of full consonants to [~] and [h], segments with only laryngeal features. Stops
in Yucatec are aspirated; here, as in Clements' examples, deletion of all supralaryngeal
features yields [h].
This presents a problem for a theory where the values are ordered, because
again the context is on the right-hand (fricative) side of the affricate, and
yet the affricate acts like the stops. But if the values of [cont] are underly-
ingly unordered, this rule is simply:
(9) [-cont] ~ [-voice]/ ]~
Since affricates are [-cont], they will undergo this rule.
3.3. Zoque Voicing and Nasal-Consonant Clusters

A rule in Zoque (Wonderly 1951) voices a noncontinuant after a nasal,
and it applies to affricates as well as stops. Fricatives after a nasal are not
affected by this rule; they remain voiceless, and in some cases (to be
discussed below) the nasal deletes.
(10) Zoque
a. Stops and affricates:
/min + pa/ [minba] 'he comes'
/min + tam/ [mindam] 'come! (pl.)'
/pAn + 6Aki/ [pAn)'Aki] 'figure of a man'
/N + pama/ [mbama] 'my clothing'
/N + 6o~ngoya/ [fi]'Ongoya] 'my rabbit'
b. Fricatives:
[winsa'~u] 'he received'
[%nsis] 'lips'
/N + s A k / ~ [sAk] 'my beans'
c. [ - cont] ~ [ + voice]l[ + n a s a l ] . - -
This rule is presented by Sagey (1986) as an example of an edge effect:

the context is on the left side of the affricate, and the affricate is treated
as [-cont]. But since this rule applies to all [-cont] segments it will
necessarily apply to stops and affricates and not to fricatives. Ordering of
the values of [cont] is not necessary to the analysis. It is also possible that
this is actually a rule that applies to any segment after a nasal, and it is
blocked from applying to the fricative by structure preservation, since
there are no voiced fricatives in the language. Under either analysis, this
process gives no evidence for a phonological edge effect.
Wonderly's description of Zoque also includes a nasal morphophoneme
that he transcribes as N. N appears as a nasal homorganic to a following
386 LINDA LOMBARDI
stop, as [n] before [w, y, h], and as zero elsewhere. The following examples
are of words in combination with a pronominal prefix N- . (Voiceless
stops after nasals become voiced by the rule in (lOc).)
(11) Zoque
N + pama [mbama] 'my clothing'
N + tatah [ndatah] 'my father'
N + kayu [ngayu] 'my horse'
N + gayu [ngayu] 'my rooster'
N + faha [faha] 'my belt'
N + s^k [s^k] 'my beans'
N + sapun [sapun] 'my soap'
N + lawus [lawus] 'my nail'
N + 6o~ngoya [fijo~ngoya] 'my rabbit'
N + cima [ncima] 'my calabash'
On superficial examination, this looks like an example of an edge effect
that would be a counterexample to the theory proposed in this paper: the
context of a rule of nasal deletion is [+cont], and the nasal is on the
[-cont] edge of the affricate, so that the rule does not apply before
affricates. However, such an analysis would be missing a wider generaliza-
tion about the behavior of nasals and [+/-cont] segments in clusters.
Compare the Zoque N to the behavior of [n] in Lithuanian. In Lithuan-
ian, [n] assimilates in place to a following stop or affricate. It deletes
before any consonant from the set [j,v,s,z,~,~,l,r,m,n]. Note that except
for [m,n], these are all the segments that are not [-cont]. The following
examples with the prefix san- exemplify this. (Examples, from Kenstowicz
(1972), are given in Lithuanian orthography - the hook on the voweI
indicates length, not nasality; accents omitted; nk sequences indicate
angma, according to his rule, although transcriptions are not given.)
(12) Lithuanian
samburis 'assembly'
sampilas 'stock, store'
sandora 'covenant'
santaka 'confluence'
sankaba 'clamp'
s~junga 'union'
s~voka 'idea'
sgskambis 'harmony'
s~glavos 'sweepings'
(12) s~ine 'conscience'

sglytis 'clash'
sorasas 'list'
s~mokslas 'conspiracy'
s~narys 'joint'
What appears to be happening in Lithuanian is that when [n] precedes
another consonant, it must share a Place node with that consonant. This
can be expressed as a condition on syllable well-formedness, a Coda
Condition (It6 1986), as in (13) ([cor] must be included because none of
these rules applies to [m]):
(13) *CI~
I
+nasal]
cot J
The nasal assimilation rule assimilates [n] to a following [-cont] segment.

This will apply when In] precedes a stop or an affricate, and will result in
a nasal-consonant cluster that is linked to one Place node:
(14) n b n b m b
[ assimilation:
"",,] = NN~
lab lab lab
Such clusters will not meet the structural description of the filter in (13),
due to the Linking Condition (Hayes 1986). Thus, a syllable-final In] that
is followed by a stop or affricate will not be ruled out by the coda
condition.
When a nasal stands before a fricative, the rule of nasal assimilation
will not apply, because fricatives are not [-cont]. Thus, a nasal standing
before a fricative will not come to share a Place node with the-following
consonant. In such cases, then, the coda condition will be violated, and the
nasal will delete. (In Lithuanian the preceding vowel then compensatorily
lengthens.) The crucial point is that the nasal deletion does NOT come
about through the action of a rule, "delete nasal before [+cont]," which
is a rule that we would expect to apply before affricates as well. It deletes
because it violates the syllable well-formedness condition above in (13).
In fact there is no rule of nasal deletion as such; deletion is simply the
result of stray erasure, which deletes all unsyllabified elements. The nasal
does not delete before [-cont] segments (stops and affricates) because
due to the action of the nasal assimilation rule, a structure is created that
does not violate the condition in (13); thus they can be syllabified, and
388 LINDA LOMBARDI
are not subject to stray erasure, s As predicted by the condition in (13),

word-final [n] deletes as well.
Before another nasal, [n] also deletes. Presumably nasals assimilate in
Place to a following nasal, since nasals are [-cont], and the result is then
subject to degemination, since there are no geminates in the language.
Thus this deletion is separate from the deletion of fricatives, but both are
the result of well-formedness conditions in the language, rather than the
result of a specific rule of deletion.
The Zoque N morphophoneme is subject to the same restriction as
Lithuanian [n]. It must be homorganic to a following consonant, and the
rule that assimilates it only applies if the following consonant is [-cont].
This is not an edge effect, then, as it seems to be superficially. The
rule of nasal assimilation applies before all [-cont] segments, stops and
affricates. Nasals that stand before [+cont] segments then delete because
the condition on well-formedness is violated. The process of deletion is
not actually a rule that makes reference to [+cont], and thus is not a rule
that would be expected to apply before affricates.
This seems to be a rather common situation cross-linguistically; for
example, Portuguese and Polish appear to require a similar analysis, with
vowel nasalization resulting when the nasals delete. (Polish does have
some nonhomorganic nasal-stop clusters on the surface, but these arise
from a later rule of vowel deletion.) In Kikuyu (Barlow 1951), [n] assimi-
lates in Place to a following [-cont], resulting in a prenasalized stop, and
deletes elsewhere (in this case, before [0, h]). Again, the nasal also deletes
before another nasal, presumably due to an independent principle. The
prediction that this analysis makes is that languages that have a rule of
nasal deletion before fricatives should always be languages that have a
rule assimilating nasals to a following [-cont] segment. 9
3.4. Modern Yucatec Maya Morpheme Structure Constraints

The native Yucatec Maya vocabulary consists mainly of monosyllabic CVC
roots (Straight 1976). There are several constraints on the cooccurence of
s A nasal will sometimes stand before a fricative that happens to have the same Place
feature, as when coronal [n] is prefixed to coronal [s] in the examples above. These nasals
do not escape the coda condition; thus they must not be doubly linked to the same Place
node, but must have adjacent identical Place nodes. This would appear to be a situation
where the OCP is not enforced. See Rubach (1986) for a proposal that the OCP does not
operate on the Place node in Polish, another language that has essentially the same facts
about nasal deletion before fricatives.
9 A stronger hypothesis also suggests itself: that all deletion rules are the result of stray
erasure - that there are no rules that delete consonants, but rather that consonants that
delete are always those that are unsyllabifiable.
consonants in a root. One of these is stated in (15):

(15) If both consonants in a root are [+cont], they must be identical.
This constraint applies to both affricates and fricatives, which are both
[+cont]. Thus, the only possible CVC roots containing only fricatives and
affricates are those given in (16a); the roots in (16b) do not occur. 1°
(16)a. cVc ~V~ sVs gVg
b.
cV~ 6Vc sVc gVc
cVs ~Vs sV6 gV6
cVg ~Vg sVg gVs
The cooccurence restrictions hold regardless of the order of the consonants
in the root; both the order/cs/, in which on the surface the [+cont] part
of the affricate is adjacent to the fricative, and the order/sc/, in which
the surface [-cont] part of the affricate is next to the fricative, are ruled
out. Vowels and consonants are transparent with respect to one another
in these roots, and it can be shown (McCarthy 1989) that they are repre-
sented on separate planes. Therefore, the consonants are effectively adjac-
ent in underlying representation.
These constraints can be analyzed as resulting from the interaction of
the OCP and a language-specific constraint against branching association
to a particular feature (McCarthy 1985, Mester 1986), as follows:
1. The OCP prohibits adjacent [+cont] specifications in these roots.
Thus the representations in (17) are ruled out, since given that the values
of [cont] are on separate tiers, the [+cont] specifications in (17a) and in
(17b) are adjacent:
(17)a. (17)b.
¢ C C
R o ~
[+cont] I
[+con~ [+con~ [+cont]
J
[-cont] [-cont] [-cont]
10 This is a slight oversimplification: glottalized affricates do not obey this constraint. I will
leave this issue aside, as it does not affect the main argument of this section. Section 3.5
gives a more detailed analysis of similar facts in Classical Yucatec. I will omit a more detailed
analysis of m o d e r n Yucatec because Straight (1976) gives only a list of the shapes of
impossible Roots and I have been unable to verify his claims with any actual lexical material.
The only dialects for which Root lists are available have slightly different MSCs. It is possible
that some of the impossible Roots are also ruled out by a requirement of anterior h a r m o n y ,
which holds in Classical Yucatec (see section 3.5) and in the other dialects. Straight lists Ix]
and [f] as p h o n e m e s of Yucatec. He states that [x] occurs only in loans; he says nothing
about [f], but I p r e s u m e that this is an oversight and that he found [f] only in loans also,
since no other sources on Yucatec list this as a sound of the language.
390 LINDA LOMBARDI
2. The language has a prohibition against a branching [+cont]. Thus

the following alternative representations for the roots in (18) are also
ruled out:
(18)a. c c b. c s
[+cont] [-cont] [-cont] [+cont]
The only possible roots that have two [+cont] segments, then, are those
that have a branching Root node, and thus are two identical consonants:
(19)a. s s b. C C
La/
I
[+cont]
[+coU
[-cont]
If the values of [cont] in the affricate were underlyingly ordered, then
only certain orderings of affricates and fricatives would violate the con-
straints:/cs/would be ruled out, because the [+cont] part of the affricate
is adjacent to the fricative (20a), b u t / s c / w o u l d not be ruled out, as the
values of [+cont] would not be adjacent (20b). This result is incorrect
since both roots are impossible.
(20)a. c s b. s e
[-cont] [+cont] [+cont] [+cont] [-cont] [+cont]
3.5. Morpheme Structure Constraints in Classical Yucatec

Root morphemes in Classical Yucatec are also of the form CVC. Appendix
1 gives all of the possible combinations of the stop, fricative and affricate
consonants of Yucatec (glides and sonorants are omitted), n If there were
no restrictions on the cooccurence of consonants, each box in the table
could contain one or more possible roots of Yucatec in which the two
consonants in the box appeared with one of the vowels of the language.
But certain of the logically possible consonant combinations do not occur.
11 Classical Yucatec was spoken in the Yucatan peninsula of Mexico from the mid-15th to
the mid-17th century. M c Q u o w n (1967) contains tables of all the roots from the Motul
dictionary, which was compiled in the last quarter of the 16th century. These charts were
the source for the table in the Appendix, checked against three other sources: the Motul
dictionary itself, Swadesh et. al. (1970), and the Diccionario Maya Cordemex.
These gaps are not accidental, and can be accounted for by considering
the underlying representation of roots in the lexicon.
The constraints can be briefly stated as follows:
1. Roots are subject to a requirement of anterior harmony: two coronals
in a root must agree in the value o f [anterior]. For example, sVc' is a
possible root, but not *sq/c', *sVC.
2. T h e r e is a restriction on the order of affricates with respect to stops
and fricatives in a root that is made up of only nonglottalized coronal
consonants. If such a root contains a stop and an affricate, or a fricative
and an affricate, the affricate always occupies the second consonant pos-
ition. Thus, tVc and sVc are possible roots, but cVt and cVs do not occur.
This constraint does not apply to glottalized affricates; tVc' and c'Vt are
both possible, as are sVc' and c'Vs. It does not apply if the consonants
differ in Place.
3. If both consonants in a root are glottalized, they must be identical.
There are roots of the form C ' I V C ' I , but roots of the form C'IVC'2 are
impossible. For example, p'Vp' is a possible root of Yucatec, but /p'/
cannot cooccur with any other glottalized consonant: *p'Vt', *p'Vk', etc.,
are impossible.
4. In Modern Yucatec, the plain and glottalized version of a consonant
cannot cooccur in a root. Classical Yucatec appears to have a restriction
on ordering rather than an absolute prohibition on cooccurrence in cases
like this. W h e r e C1 = C2, C V C ' is impossible, but there are a few examples
of the f o r m C'VC. Although the latter sort of example is rare, there are
absolutely no examples of the former type.
In the present instance, constraint 2 in Classical Yucatec is of particular
interest. This constraint provides additional evidence that the underlying
representation of the affricate must not be an ordered representation, as
the following analysis will demonstrate.
3.5.1. Constraint on Ordering of Affricates: the Lexical Entries of These

Roots in Yucatec
Considering only nonglottalized coronal consonants, the following con-

straints hold on the ordering of an affricate with respect to a stop or a
fricative in a root: a stop and an affricate can cooccur in a Yucatec root,
but only in that order. This means that roots such as tVc are possible, but
roots such as *cVt are not. Fricatives and affricates can cooccur as well,
but again, only in that order: sVc is possible, but not *cVs.
Thus, the ordering of the stop or fricative with respect to the affricate
is never contrastive in these roots. If we are to eliminate redundancy in
392 LINDA LOMBARDI
lexical entries, including redundancies in linear order (McCarthy 1989),

the representation of these roots should not permit a contrast between,
for example, the order stop-affricate and the order affricate-stop, since
the latter order is impossible.
Affricates have two values of [cont]; stops and fricatives each have only
one. In a root that contains a stop and an affricate, both are [-cont]; in
addition, the affricate is specified [+cont]. Thus the lexical entry must
contain the feature [ + cont], which distinguishes the stop and the affricate.
This [+cont] appears on the second C in all cases, since the order affricate-
stop never occurs.
Although the feature [+cont] is linked to the second C on the surface, if
[+cont] is linked to the second C in the lexical entry, such a representation
predicts that the linking of this feature is contrastive, since it is specified
in the lexicon. It predicts that the feature could be linked to the first C
as well. This is incorrect, since such a linking never occurs; the first
consonant is never an affricate.
Thus, under the assumption that the lexicon should not contain redun-
dant information, but only information that is contrastive, in the underly-
ing representation of these roots [+cont] should not be linked to a root
node of a consonant. The linking of the feature is predictable, so it should
not be specified in the lexical entry; the feature can associate to the
rightmost C by a later rule.
3.5.2. Derivations
Take a root of the form tVc, which consists of a stop and an affricate. As
in Modern Yucatec, vowels and consonants are represented separately in
the lexical entries of these roots, since their ordering is predictable. So
we need consider only the representation of the consonants. In a root such
as tVt, the underlying representation will consist of only one consonant;
association to the root template will result in a root with two identical
consonantsl The lexical entry will be j u s t / t / p l u s a separate specification
for the vowel, or roughly as in (21) (irrelevant structure omitted):
(21)
ont]
Place
When this entry is associated to the CVC template, the single Root node
will spread to fill both positions, which are obligatory.
(22) C V C
••[-cont]
[coronal]
= tVt
A stop-affricate root such as tVc will also be represented as a single

consonant. It will consist of one Root node with features; however, the
entry also contains the feature [+cont], unassociated to the root node:
(23) Root / ~ , , ~ [+cont] (unassociated)

f [-cont]
[coronal]
Again, the Root node will spread to fill both C positions in the template.
Alone, this process would yield the same root as the entry above: tVt.
But this entry contains additional material that distinguishes it from that
entry: the unassociated [+cont]. Since it is unassociated, it must be linked
by rule; the rule is given in (24):
(24) Link unassociated features to the rightmost possible C.
This rule will link any unassociated feature, and thus will apply to both
values of [cont] (more precisely, to both features, [stop] and [cont].) 12
Various solutions have recently been proposed for cases such as this,
where a rule must have the effect of linking a feature by referring to a
specific skeletal position rather than a Root node. (See Hayes 1988, Ster-
iade forthcoming, Selkirk 1988.) The particular solution that is adopted
is not important to the present argument; I will assume the following:
when the association required by the rule takes place, the structure of the
root will change minimally to allow association to the C, as in step 2 of
the following derivation. ~3 Thus, in the output of the rule that associates
the unassociated feature, the two consonants of the root will have their
own separate Root nodes.
(25)a. Root node spreads to fulfill template:

C V C
-cont] [+cont]
[coronal]
12 Floating Laryngeal features will be linked by a separate rule; Constraint 3 shows that
floating Laryngeal features are linked to the leftmost consonant.
13 This idea is due to John McCarthy.
394 LINDA LOMBARDI
(25)b. Associate unlinked feature to the right, changing structure min-

imally as necessary:
C V C
,>/.~~
[coronal]
[+cont]
This will yield the root tVc. It is correctly predicted that the root *cVt
will never occur, since the rule will never link the floating value to the
first consonant position.
In Step 2 of the derivation, the consonants have two Root nodes, since
this is the minimal change in structure that will make it possible to link
the unassociated feature to the second consonant. If the feature were to
link to the single Root node of Step 1, it would yield a Root with two
identical affricates. Such roots are possible, but they are represented
underlyingly with two values of [cont] linked to a single Root node; there
is no reason to represent them with an unassociated feature.
Fricative-affricate roots such as sVc will have exactly the same kind
of representation except that the unassociated feature is [-cont]. The
underlying representation consists of one Root node specified [ + cont], and
an unassociated feature [-cont]. The association of the [-cont] feature is
not contrastive, so it is not specified in the lexical entry; it attaches by
rule to the rightmost C. As above, roots with the affricate in the first C slot,
such as cVs, are correctly predicted to never occur, since the unassociated
feature links by rule to the second C.
(26)a. Root node spreads to fulfill template:

C V C
--~[+cont] [-cont]
[coronal]
b. Associate unlinked feature to the right:

C V C
,, _.,.,"~'~ ~" [-cont]

,,~-'"- [+cont]
[coronal]
3.5.3 The Place and Laryngeal Node in These Roots
Constraint 2 only applies to coronals; also, it does not apply when the
affricate is glottalized. Moreover, if the place of articulation is not the
same, affricate/stop and affricate/fricative roots can occur in either order;
for example, all of the following are possible:
(27) kVc cVk sVc cVs

If the affricate is glottalized, again, both orders are possible:
(28) t V c' c' V t s V c' c'Vs
This constraint on ordering results from the form of the underlying repre-
sentation of these roots, combined with the directionality of the linking
of unassociated features. The Cs in coronal-only, nonglottalized roots
share all features except for the feature that is unassociated. The latter
feature must be unassociated because of the fact that its association is
predictable and not constrastive. When this feature is not associated, the
two consonants in the root share all remaining features. Because of the
OCP, the only option for representing them, then, is as a single Root
node; this node will fill both Cs in the template. Whatever value of [cont]
is linked to this Root node will of course appear on the C in both positions;
the unassociated value of [cont] will be linked by a rule.
Roots containing Cs that differ in place are predicted by this analysis
to have no constraint on ordering. If the two consonants in a root do not
share place of articulation, the two Cs in the root must have separate
Root nodes, since they need to have different Place nodes. The entry for
kVc is (29); the two consonants cannot merge to form a single Root node
since they do not share all features:
(29) k c
Root~~.....-"~ ~,,,~
[dolrsal ] [-contl [co!onal] [+cont]
If the Root nodes remain distinct, each Root node can have its own value
or values of [cont] attached to it. So any order of stops, fricatives and
affricates will be possible if the consonants differin Place.
Affricates that are glottalized will not participate in the constraint on
ordering for the same reason. The glottalized affricate does not share all
features with the other C in the root, because it has the feature [cg]. Thus
the root cannot be represented as a single Root node. The lexical entry
will contain two separate root nodes, as in (30):
396 LINDA LOMBARDI
(30) c' t
[ + c ~
[+cont] [coronal]
Again, since the Root nodes remain distinct, each can have its own value
or values of [cont], and there will be no constraint on ordering.14
3.5.4. Classical Yucatec MSCs and Contour Segment Theory
As shown in the preceding section, an unordered theory of the affricate

allows a unified analysis of the ordering constraints on affricates in Classi-
cal Yucatec. This section will show that such an account is impossible
given a contour segment representation of the affricate.
In a theory where the values of [cont] were ordered underlyingly it
would be simple to capture half of the constraint on ordering of affricates.
Given the representation in (31) and an unassociated [+cont] associating
to the right, the resulting Root will be a stop followed by an affricate.
(31) C V C C V C
[-cont] [+contl [-contl [-cont] [+cont]

A root with an affricate preceding a stop would be impossible; if the
unassociated feature associated to the first C, it would produce an imposs-
ible segment in which [+cont] precedes [-cont]. The association that
would yield an affricate in the first position is impossible, because it would
14 This analysis predicts that glottalized affricates might have a constraint on ordering with
respect to glottalized stops (or glottalized fricatives, but these do not occur in this language.)
The two consonants in such a Root would share all features except for [+cont] on the
affricate, and could be represented as a single Root node and a floating [+cont]. However,
due to another constraint, Constraint 3, two nonidentical glottalized consonants cannot
cooccur in a Root. Thus, this prediction is impossible to test.
It is also possible that glottalized affricates in this language are not underlyingly specified
for [ - c o n t ] . Since there are no glottalized fricatives, glottalized affricates could be the
realization of any segment that is specified for both [cg] and [+cont]. However, this does
not have any effect on the statement of the morpheme structure constraints. The fact that
the glottalized affricates need to have separate Root nodes will be sufficient for them not to
be affected by the ordering constraint.
THE NONLINEAR ORGANIZATION" OF THE AFFRICATE 397
result in crossing association lines, or what is, by the contour segment

hypothesis, the wrong order of the [cont] gestures.
(32)
C V C *C V C *C
[ or/,,,
[+cont] [-cont] [+cont] [-cont] [-cont] [+cont] [-cont]
Thus, it is possible to explain the ordering of stops and affricates in a

contour segment theory. However, that theory makes the wrong predic-
tion about the ordering constraint on fricatives and affricates. In the
representation in (33), the only place the feature can associate is to the
first C, because that is the only possible well-formed segment. This will
yield an affricate followed by a fricative, which is the order that is actually
prohibited in this language.
(33) C V C *C V C
[-cont] [+cont] [-cont] [+cont] [+cont]

It would be impossible to produce a representation of the order that is
actually possible - that is, fricative followed by affricate - because that
would result in line-crossing.
(34) C V C *C V C
[+cont] [-cont] [+cont] [-cont]

Thus, in the contour segment theory, we would expect the constraint to
be:
possible: stop-affricate, affricate-fricative
prohibited: *affricate-stop, *fricative-affricate
This is incorrect, as the actual situation is:
possible: stop-affricate, fricative-affricate
prohibited: *affricate-stop, *affricate-fricative
398 LINDA LOMBARDI
Thus it is impossible to achieve a unified analysis of this constraint under

the contour-segment theory. If we were to preserve the contour segment
theory of the affricate, the facts about these roots would have to be
analyzed as the result of two separate constraints: a constraint on the
ordering of stops and affricates, and a constraint on the order of fricatives
and affricates. This would fail to capture the generalization that in both
cases, an affricate must always appear in the second consonant position
in the root.
If the two values of [cont] were on separate tiers, the representation in
(34) would not cross association lines. However, it is generally presumed
that features on separate tiers in a single segment cannot be ordered
(Sagey 1986). One might attempt to salvage the contour-segment theory
of the affricate by saying that at some point in the derivation, or in
underlying representation, these roots can contain segments with unor-
dered values of [cont], and the values are ordered later by a rule. How-
ever, in order to account for the facts, the values would have to be
unordered throughout the phonology and the rule ordering them would
apply only to yield the phonetic representation. As the examples discussed
above show, the values of [cont] must be unordered at underlying form
(the morpheme structure constraints), and throughout the phonological
derivation (e.g., Zoque) including the postlexical derivation (between-
word processes such as the Yucatec and Basque rules in section 3.1).
Thus, the changes that need to be made to the contour segment theory
to account for these facts yield exactly the theory proposed in this paper
- that the values of [cont] are phonologically unordered, and only ordered
in the surface phonetic representation.
3.6. Summary
I. In Basque, stops delete before stops; affricates become fricatives in the
same environment. Although the context is on the [+cont] edge of the
affricate, nevertheless affricates pattern with stops; this rule can see the
value of [cont] that is not adjacent on the surface. A nearly identical rule
operates in Yucatec Maya (the difference, that stops become [h] rather
than deleting, is a result of independent factors.)
II. A rule in Turkish devoices all syllable-final [-cont] segments. This
rule applies to affricates, despite the fact that the context (the end of the
syllable) is on the [+ cont] edge of the affricate.
III. In Zoque, a [-cont] segment is voiced after a nasal; the rule applies
to affricates. Although the context is on the [-cont] edge of the affricate,
there is no need to assume ordering in order to state this rule; it simply

applies to all [ - c o n t ] segment~.
IV. Various languages delete nasals before fricatives and not before
affricates. This process is not due to the action of a rule "delete nasal
before [+cont]," which would be a rule that would be expected to apply
before affricates if the values of [cont] are unordered. Rather, it is the
result of the combined effects of a rule that assimilates nasals to a following
[ - c o n t ] and a condition that requires nasals to share Place with a following
consonant. Because the rule will not assimilate a nasal to a following
fricative, the nasal will delete because it violates the syllable well-for-
medness condition.
V. A morpheme structure constraint in Yucatec Maya disallows two
nonidentical [+cont] consonants in a root. This constraint applies regard-
less of the ordering of the consonants, and includes affricates. Classical
Yucatec has a morpheme structure constraint that requires that if a root
is made up of a stop and an affricate, or a fricative and an affricate, the
affricate must always occupy the second consonant position. The analysis
of both constraints requires that the values of [cont] in the affricate be
underlyingly unordered.
This evidence shows that many processes can see the nonadjacent value
of [cont] in a affricate. The evidence includes (1) rules in phonological
derivations such as the Basque rule in (I), (2) rules that operate between
words, such as the Basque and Yucatec rule in (I), which must be applying
postlexically, and (3) morpheme structure constraints, such as (IV), that
must operate on underlying representations. It is also clear that such
processes occur at different levels within the same language: the Basque
rule operates both in the phonological derivation and postlexically, and
Modern Yucatec has both morpheme structure constraints and a postlex-
ical rule. Assuming an unordered representation of the affricate at all of
these levels permits a straightforward analysis of these processes, as well
as of apparent edge effects such as the Zoque rule in (III) that mentions
the value of [cont] that happens to be adjacent on the surface.
In the next section I turn to processes that appear to show that there
are edge effects with affricates - that some rules cannot see the nonadjac-
ent value of [cont].
4. EDGE EFFECTS AND APPARENT EDGE EFFECTS
It is useful to consider what true counter evidence to the present proposal

would look like. Although the values of [cont] are unordered in underlying
representation, they are ordered on the surface. Thus, one must first of
400 LINDA LOMBARDI
all be careful to distinguish phonological and phonetic processes. Phonetic

processes, which occur after the ordering takes place, will be expected to
show edge effects. A counterexample to the hypothesis about underlying
ordering would have to be a rule that is phonological and which treats
affricates as stops on their left edge only, or as fricatives on their right
edge only.
Sagey (1986) presents examples of rules that are sensitive to the edge
of the affricate that the context appears on. One of these is the postnasal
voicing rule in Zoque discussed above in section 3.3; as shown there,
ordering is not necessary for the statement of this rule, which simply
applies to all [-cont] segments. Other examples appear more problematic
at first glance. However, as the next section will show, these examples
are actually phonetic edge effects, which are exactly what the theory
predicts.
4.1. Sierra Popoluca
The distribution of aspiration in voiceless stops in Sierra Popoluca is also
presented in Sagey (1986) as an example of an edge effect (data from
Foster and Foster (1948)). Stops are aspirated at the end of a syllable,
but affricates and fricatives are not:
(35) SierraPopoluca
a.~ stops
/hop/ [hoph] 'mouth'
Pampat/ [~ampat h] 'I met'
/rook/ [mok hI 'fog'
b. affricates
/mac/ [mac] 'grasp'
Papi~/ [~api6] 'thorn'
c. fricatives
/wosten/ [wosten] 'two'
/pi~tokh/ [pi~tak h] 'flea'
Sagey's analysis is that this is a rule applying to a [-cont] at the end of

a syllable. In an analysis where the affricate has ordered underlying values
of [cont], this rule will fail to apply to an affricate because the [+cont]
value is on the right edge, at the end of the syllable. Thus the structural
description for the rule is not met. Sagey presents this as a phonological
rule, but as predicted by the theory in this paper, this is a rule that shows
an edge effect because it is a phonetic process.
To begin, note that the facts about aspiration in Sierra Popoluca are
somewhat more complex than stated above, where only examples of word-
final stops and affricates are given. The complete description is as follows
(Elson (1947)):
1. Voiceless stops are unaspirated when followed by a vowel, or by a
consonant at the same point of articulation.
(36) /kek.gak.pa/ [kek.gakh.pa] 'it flies again'
2. Voiceless stops are aspirated when followed by a nasal, or when

syllable final, except when there is a following syllable that begins with a
consonant at the same point of articulation.
(37) /k~k.pa/ [k~kh.pa] 'it flies'
3. If the first member of a nonhomorganic consonant cluster is a nasal,

schwa is inserted between the members of the cluster.
(38)a. Pi.p~n.pa/ [~i.p~no.pa~] 'he builds a nest'
b. /min.pa/ [mino.pa:] 'he comes'
The aspiration rule is not structure-preserving; aspiration is not distinctive

in this language. Elson describes these differences in consonant clusters
in Sierra Popoluca as the difference between open and close transition.
What we have here is not a phonological rule, but a statement of the
conditions under which stops are released or unreleased in this language;
this must be accounted for by a rule of phonetic implementation. Ander-
son (1974) observes that release of consonants is never a distinctive pro-
perty within a language; it is never a property that is manipulated by
phonological rules. Therefore, cross-linguistic differences in release must
be a result of differences in phonetic implementation rules. Browman and
Goldstein (1986) propose a theory of phonetic implementation rules that
explains such cross-linguistic variation as differences in relative timing of
articulatory gestures in the phonetic component. Levin (1987) provides a
theory of these release effects, which she interprets as "excrescent vow-
els". Levin shows that these excrescent vowels must be inserted in the
phonetic component, in contrast to epenthetic vowels, which are inserted
in the phonology. Because Levin's proposal is worked out in some detail,
I will compare Sierra Popoluca to Levin's example of Piro, to demonstrate
that the effect in Sierra Popoluca is also a phonetic effect.
Excrescence is not triggered by stray consonants, but rather seems to
be required as a transition between adjacent articulations. This appears
to be the function of this rule. If the adjacent articulations are identical,
no insertion occurs. Clements (1985) suggests that this is due to a rule
402 LINDA LOMBARDI
merging adjacent identical Place nodes, to avoid a violation of the OCP;

in any case, there is basically only one articulatory gesture. I n nonhomor-
ganic clusters, there is a transition from one articulation to another (what
Elson describes as " o p e n transition"), and the language requires particular
ways o f making the transition.
Under the principles of underspecification theory, an epenthetic vowel
should be the default underlying vowel of the language. An excrescent
vowel can be a vowel that is not underlying: as in Levin's example of Piro,
where the excrescent vowel is schwa. The excrescent vowel in Popoluca is
also schwa, a segment that Elson does not consider to be a phoneme of
the language.
Also, in Piro, in some cases rather than an excrescent vowel being
inserted, a consonant may become syllabic: a sonorant or continuant
becomes syllabic if it is greater in sonority than the following consonant.
In Popoluca nonhomorganic clusters, the effect of the rule is different for
stops, nasals, and affricates. Nasals take an excrescent vowel. Stops be-
come aspirated, which can be seen as having the effect of making the
stop syllabic, since in languages with syllabic consonants aspiration may
function as the syllable peakl In fact, in Piro, whether the excrescent
vowel is voiced or voiceless depends on the context; possibly the aspiration
in Popoluca is basically the same, a voiceless vowel appearing adjacent to
voiceless segments, while schwa, a voiced vowel, appears with the nasals,
which are voiced. (/g/is the only voiced stop that is common in Popoluca,
and it never occurs syllable-finally before a C-initial syllable, so it will
never participate in this process). And in Popoluca it appears that the
fricative is sonorant enough that no insertion is needed when the first
m e m b e r of the cluster is a fricative or affricate.
The facts about this phonetic process in Sierra Popoluca are similar to
the description of the lexical syllabification process in Bella Coola by
H o a r d (1978). In the case of stops, aspiration is the syllable peak; fricatives
can be syllabic on their own, so frication is a possible syllable peak. Bella
Coola differs in that affricates take aspiration as their syllable peak, rather
than the fricative portion of the affricate. However, in fact it is also
possible that Popoluca affricates are aspirated and are just not so tran-
scribed. Affricates in English, for instance, show similar voice onset time
as stops, but are not usually transcribed as aspirated. 15
15 Archangeli and Pulleyblank (1987) present data from Nahuatl that give evidence of the
same phenomenon as in Sierra Popoluca: stops are aspirated syllable-finally,but not affricates
or fricatives. I was unable to obtain any more data on this phenomenon in Nahuatl, but it
seems probable that the explanation is similar to that for Sierra Popoluca.
4.2. Kutep
Labialization in Kutep is presented by Sagey (1986) as an example of an
edge effect. Labialization results in a labiodental [f] or [v] after fricatives
and affricates, and a bilabial [w] after stops.
(39) Kutep
a. Fricatives:
basra 'they kneel'
nsazVakkWa 'the water is hot'
ba~Ve 'they washed'
ba~vam 'they begged'
acfapan 'groundnuts'
b. Affricates:
batsrap 'they chose'
bat9fak 'they sleep'
c. Stops:
bapwa 'they grind'
bambWa 'they tasted'
bandWap 'they wove'
nsazVakkaWa 'the water is hot'
bangWa 'they drink'
baskWap 'they are foolish'
There is clearly a difference between the stops and the affricates. How-
ever, there is no evidence that this is a phonological process or a rule of
any kind. The original source for these data is a list of the phonemic
contrasts in this language, and the source, Ladefoged (1968), gives no
other information about the language. For example,/basfa/contrasts with
Poasa/, 'they took'; this is a minimal pair, not the effect of a phonological
rule. The difference between [f]/[v] and [w] appears to be merely how
this language realizes the feature [round]. Rounding the lips when a
fricative is being produced will presumably give the effect of [f] or [v]
depending on the voicing of the fricative. Rounding of the lips during a
stop gives the effect of [w] when the stop is released. Since affricates end
phonetically in a fricative release, the effect of rounding on a fricative will
be heard after an affricate, not the effect of rounding on the release of a
stop.
404 LINDA LOMBARDI
Compare this to labialization in Higi, a Nigerian language. Mohrlang

(1972) discusses the differences in phonetic implementation of labialization
with different classes of consonants. Preceding stops and affricates, "the
lips tend to completely close, and the effect is that of a rounded bilabial
preceding the consonant". Preceding fricatives, "the lips do not com-
pletely close. The e f f e c t . . , is that of a rounded bilabial fricative preceding
the consonant". Some examples are given in (40).
(40) Higi
a. Stops:
/wt] [Pt~i] 'skin'
/wd/ [bdcni] 'to pour'
b. Affricates:
/Wts/ [Ptsi] 'grass'
]Wdz/[bdzi] 'strand'
c. Fricatives:
/Ws/ [Wsq 'thing'
/Wz/ [Wza] 'farming'
This looks like a case of an edge effect: the labialization seems to occur
on the left-hand side of the affricate, which is the [-cont] side, and the
affricates are patterning with the stops. But once again, there is no reason
to consider this a phonological rule sensitive to edges. Effects like this
and the one in Kutep are the result of the phonetic realization of certain
combinations of features, and have no bearing on the ordering of features
in underlying representation. In both cases, what is at issue is the differ-
ence in phonetic implementation of a phonological contrast (a distinctive
feature) in the two languages. Since the values of [cont] are ordered
phonetically, the phonetic implementation of labialization will show effects
of this ordering.
4.3. English
Borowsky (1987) presents an analysis of the plural and past tense endings
in English which depends crucially on the ordering of the values of [cont]
in the affricate. The operation of the deletion and voicing assimilation
rules in (41a,b) is shown in (41c):
(41)a. Inflectional Vowel Deletion

V ~ O/.--C
e~cont
[ oo -]
b. Progressive Voicing Assimilation

[+voice] ~ [ - v o i c e ] / [ - v o i c e ] . _ _
[ ....c o t ,]
c. book + Vz love + Vd bee + Vz
book z love d bee z Deletion
books -- -- Voicing
assimilation
[bUks] [1Avd] [biyz]
Deletion of the vowel in either morpheme is blocked when an OCP
violation would result. (The analysis is stated in terms of the deletion of
the vowel in the ending and the conditions under which such deletion is
blocked, but the issues are exactly the same for an analysis in which the
vowel is epenthetic.) For instance, deletion is blocked when two adjacent
identical segments would result:
(42) roz + Vz *rozz

p~ed + Vd *pa~dd
Deletion is not only blocked when identical segments would result,

however. It is blocked when the segments differ only in voicing, so that
deletion does not create the sequences *[td], *[sz]. It is also blocked when
the plural ending is added to any affricate or fricative: churches, judges,
bushes. To account for this, it is proposed that the OCP examines the
Manner node and the Primary Place feature in a feature geometry like
that in (43); if these are identical, deletion does not occur. This will rule
out deletion which creates [td], [sz], [gz].
(43) g z
I I
[+cont] [+cont] Manner node.]
, , ;,:Cma;de]' Identity for

the OCP
[-ant] [ + ant] Secondary

406 LINDA LOMBARDI
It is assumed that affricates are ordered contour segments, and that dele-
tion will be blocked when the plural morpheme is added to an affricate,
since it would result in two adjacent segments with the same primary Place
features and adjacent [+cont] features. When the past tense ending is
added to a coronal stop, again, deletion would result in adjacent segments
identical in primary Place and Manner, and so it will be blocked.
If affricates are unordered, this analysis predicts that deletion would
also be blocked when the past tense ending is added to an affricate: it
would result in adjacent segments which have the same primary Place
features, and adjacent [-cont] values. But this prediction is incorrect:
deletion does apply in, for instance, watched and fudged. If this is the
correct analysis of the English plural and past tense, this would appear to
be evidence that the values of [cont] must be ordered in the phonological
derivation. However, there are a number of problems with this analysis
of the conditions on deletion. As I will show, a solution to these problems
also eliminates the apparent problem that this presents for the unordered
representation of affricates.
The analysis makes several problematic assumptions which differ from
well-supported current versions of feature geometry. There is no Primary
and Secondary Place - rather, [anterior] is a dependant of the feature
Coronal (Sagey 1986). There is also no evidence for a Manner node (Sagey
1986, McCarthy 1988). (In (43) Place is also dependant on the Manner,
node, but this does not appear to play any real role.) The analysis ignores
the fact that in the plural ending, deletion is only blocked if the [+cont]
segment ending the word is also [+strident]; the vowel is deleted in, for
example, paths and leaves. This could possibly be handled by placing
[strident] in the Manner node, although if the feature [strident] exists (see
Keating 1988) it may be restricted to coronals and possibly should be
dependent on the Coronal node. The hypothetical Manner node would
presumably also have to contain [sonorant], or deletion would be blocked
in gunned, cars, pills.
But even if there were independent evidence for these modifications to
the feature geometry, the rule cannot actually be stated as applying to the
abstract nodes. One cannot talk about identical Place nodes if the entire
node is not identical; thus, the condition cannot be stated as occurring
when the Place nodes are identical, because deletion occurs even when
the segments differ in values of [anterior]. Even assuming that there is a
Manner node and that it contains [cont], [strident] and [sonorant], it will
still be impossible to state the rule as applying when the Manner nodes
are identical. The Manner node of the affricate is branching, and thus will
not be identical to the Manner node of the fricative; yet deletion must be
blocked when the plural morpheme is added to an affricate-final stem.

Thus the analysis not only assumes an unorthodox version of feature
geometry, but is not actually using any of the properties of the feature
hierarchy: although stated in terms of identity of nodes, the Place and
Manner nodes, it really needs to be stated in terms of identity of adjacent
feature values. Since the rule cannot be stated in terms of nodes, it is
actually a very complex rule: it must block deletion when adjacent seg-
ments have identical values of [cont], [strident], [sonorant], and major
articulator feature.
The analysis attempts to capture the insight that principles of well-
formedness in English rule out deletion in these morphemes. It assumes
that the same principle rules out deletion in the case of the plural and the
past tense. The latter assumption appears to be the cause of the difficulties
in formalizing the analysis, and I will modify the analysis in light of
these considerations, retaining Borowsky's essential insight that deletion is
blocked when a well-formedness condition would be violated.
The past tense ending is a coronal stop: deletion is blocked when the
morpheme is added to a stem ending in a coronal stop. This is due to a
general constraint on English. English prohibits geminates; also prohibited
are two adjacent consonants which differ only in their value for the feature
[voice], such as *[td]. This can be ruled out by the OCP if the OCP not
only rules out adjacent segments which are identical, but those which
are nondistinct (as argued by Yip (1989) following a suggestion by John
McCarthy). If only voiced consonants have a value for [voice], either
through underspecification or due to [voice] being a privative feature,
segments which differ only with respect to voicing will be nondistinct
(Chomsky and Halle 1968, p. 382). The OCP will then rule out deletion
which results in the sequence *[td].
In English, the OCP also prohibits two adjacent tautosyllabic [strident]
segments. Given an unordered representation of the affricate, this con-
dition accounts for the form of the English plural morpheme and for other
facts as well. For instance, initial clusters of [s][-cont] are permitted in
English: [st, sk, sp]. In the contour segment theory, we would expect that
[s]-affricate clusters would be permitted, too, since the [s] is adjacent to
the [-cont] edge of the affricate. But these are impossible. The condition
against two adjacent [strident] segments accounts for this fact. This con-
dition also happens to be invoked in some cases of pluralization because
the plural ending is itself [strident]: in those cases where it is added to a
stem which is [strident], whether affricate or fricative, deletion will be
blocked. Thus, this constraint accounts for the absence of both *[~s] and
*Is6] where they would otherwise be expected.
408 LINDA LOMBARDI
Given the above account of deletion blocking with the English plural,
deletion will not be blocked when the past tense morpheme is added to
an affricate-final stem, since this would not create a geminate nor violate
any other constraint. The sequences [jd], as in judged, is not a sequence
of two nondistinct consonants, nor does it have two adjacent [strident]
segments.
The conditions on vowel deletion in the plural and past tense mor-
phemes are a result of general conditions on well-formedness in English,
and these conditions may both be OCP effects, but this does not mean
that the same effect is operating in the two cases. The correct generaliza-
tions are those which relate the restrictions on deletion in these mor-
phemes to general properties of English, but these will not necessarily be
the same for the two morphemes. It is unlikely that any version of feature
geometry would allow one to collapse the two constraints described above.
But more important, it would be incorrect to try to collapse them, as their
domains are different. The constraint against geminates holds of adjacent
segments, whether tauto- or heterosyllabic, through the Level 1 phonol-
ogy. (Heterosyllabically, both are possible at Level 2: geminates in, for
example, subbasement; sequences which differ only in voicing in hump-
back, subplot, hotdog.) The constraint against two [strident] segments
holds only tautosyllabically; [s~] sequences are possible heterosyllabically,
as in question, exchange, disjunct.
The behavior of affricates in these processes is consistent with the
unordered representation. Deletion occurs when the past tense morpheme
is added to an affricate-final stem, because no constraints are violated;
deletion does not occur when the plural ending is added to an affricate-
final stem, because the constraint on adjacent [strident] is violated. In
addition, this restatement of the constraints on deletion also allows an
account of the absence ofi--nitial--*[s~] clusters, which would be expected
to occur under the contour segment theory.
4.4. Summary
The theory presented in this paper predicts that phonological processes
will never show edge effects. The cases examined supported this predic-
tion. However, since the values of [cont] are ordered phonetically, it is
predicted that there will be phonetic edge effects. The Sierra Popoluca and
Kutep examples were used by Sagey to argue for an ordered phonological
representation of the affricate. Closer examination shows that these are,
as predicted by the unordered theory, phonetic and not phonological.
5. OTHER PROPOSALS
Jakobson, Fant and Halle (1951) proposed that affricates are basically
stops with the additional feature [+ strident]. As pointed out by Chomsky
and Halle (1968), there are affricates that are not strident, such as Chipew-
yan [tO]. Chomsky and Halle thus represent affricates as stops with the
additional feature [delayed release]. Any theory which represents affri-
cates as basically stops with some other additional feature will have the
result that affricates will behave as [-cont] from both edges; presuming
that they are continuant on one edge in phonetic representation, phonetic
edge effects will also follow. However, this will not account for those
cases where affricates pattern with fricatives. The Yucatec morpheme
structure constraints, for example, show that affricates must be [+cont]
in underlying representation.
A theory where contour-segment affricates are distinguished from stops
in underlying representation by some other feature ([strident], [del rel],
or some new feature) could have [+cont] specified later by a default rule.
This would predict that after the value is filled in, the affricate will no
longer be treated as [-cont] on its right edge. Since we have seen that
there are rules which treat the affricate as [-cont] on its right edge as
late as the postlexical phonology, this solution will not yield the correct
results. In fact, we have seen that in a single language, Yucatec, there is
evidence from morpheme structure constraints that [+cont] must be in
underlying representations, and evidence from the rule in section 3.1 that
affricates are treated as [-cont] on the right edge later, in the postlexical
phonology.
In an autosegmental framework, Clements and Keyser (1983) represent
the affricate as two segments linked to a single skeletal position. As
pointed out by McCarthy and Prince (1990), this cannot be correct, since
affricates behave as single units melodically. If affricates were only joined
at the skeleton, it would be impossible to have underlying affricates in a
language where melody and skeleton are not linked in underlying repre-
sentation, such as in Yucatec and Arabic; yet both languages have affri-
cates. There is also no explanation for the fact that all features of an
affricate, are shared aside from the values of [cont]; for instance, that the
two segments of an affricate do not differ in Place. (This representation
is impossible, of course, if the skeleton is not made up of segment-sized
units, as proposed by McCarthy and Prince (1990).)
Kaye (1985) argues that short diphthongs consist of two unordered
elements linked to a single skeletal slot, their relative order being deter-
mined by a general principle that such a representation is realized with
410 LINDA LOMBARDI
the least sonorous element first. He alludes to extending this idea to

affricates, as well as prenasalized stops, although an exception to the
ordering principle must be made for prenasalized stops. This appears
to be the first suggestion that affricates are unordered; otherwise the
representation is essentially identical to that of Clements and Keyser, and
the same objections hold.
Sagey (1986) represents the affricate as two ordered values of [cont]
contained within a single segment, as discussed above in the Introduction.
This proposal has some clear advantages over earlier proposals. It accounts
for the fact that affricates are single units melodically, and pattern some-
times with stops and sometimes with fricatives. However, as I have shown,
the stipulation that the values are ordered presents problems for the
analysis of various processes that Sagey does not discuss. In the remainder
of this section I discuss recent proposals to revise the representation of
the affricate in a way that will handle this problem, and compare them to
the theory presented in this paper.
5.1. Maximal and Minimal Scansion with an Ordered Affricate
Anti-edge effects with affricates were first noted by Archangeli and Pulley-
blank (1987). They treat this problem in the context of a general approach
to the issue of phonological locality. Archangeli and Pulleyblank introduce
the idea that phonological rules can involve either maximal or minimal
scansion, which is a parameter set for each rule. Rules involving maximal
scansion scan the segment from the level of the skeleton. This means that
the rule can see all features of the segment that are below the skeletal
level. Thus, such rules will have access to both values of [cont] in an
affricate. Rules involving minimal scansion scan only the tier containing
the feature in question. Such rules will show edge effects; they are only
able to see the value of [cont] that is immediately adjacent to the context.
Thus, the rule in Basque discussed in section 3.1 and the Turkish rule
in section 3.2 are analyzed as cases of maximal scansion. The Basque rule
is stated as:
(44) [-cont, -son] --~ 0/ [-cont]
This rule must have a parameter set for maximal scansion: the rule scans
the representation from the level of the skeleton, from which it can see
the nonadjacent, [-cont] feature value. Affricates are represented as
having two Supralaryngeal nodes, and the different effect of the rule on
stops and affricates is said to result from the fact that the Supralaryngeal
node will be deleted by this rule. This will result in the loss of all features
if the segment is a stop, but will leave behind all the features of a fricative
in the case of an affricate. The fate of the Root node of the stop after
the deletion of its Supralaryngeal node is not explicitly discussed, but
presumably it deletes because it does not constitute a well-formed seg-
ment.
This analysis assumes ordered values of [cont] and accounts for non-
edge effects by the parameter that sets the difference in scansion. As I
have shown, these results are achieved without additional stipulation (such
as that involved in the scansion-setting parameter) if there is no ordering.
More important, there are cases that cannot be accounted for by a differ-
ence in scansion, such as the modern Yucatec morpheme structure con-
straints discussed in section 3.4.
As McCarthy (1989) demonstrates, the rigid CVC shape of native Yuca-
tec roots makes both this shape and the relative ordering of vowels with
respect to consonants predictable; thus they should not be part of the
lexical representations. There is, for example, no possible contrast among
the roots like /tka/, /tak/, and /atk/, in Yucatec; only /tak/ is possible.
Thus, if redundancy is to be eliminated from the lexicon, the only possible
lexical entry for the root is/a/,/tk/, with separate representation of vowels
and consonants and no underlying association to the skeleton. The conso-
nants are adjacent in underlying representation, as is required to explain
their behavior with respect to the morpheme structure constraints. The
vowel is not ordered with respect to the consonants, because the rigid
CVC shape insures that ordering. Most important to the present point,
this CVC skeleton is also not part of the underlying representation, since
all Yucatec roots are of this form.
This being the case, the morpheme structure constraints, which hold at
the level of underlying representation, must be operative at a level where
there is no CVC skeleton. But maximal rules are the only rules that show
anti-edge effects, since by scanning from the skeleton, these rules can
see both values of [cont]. Consequently, because there is no skeleton in
the underlying representation of the Yucatec morphemes, the constraints
cannot involve maximal scansion. In the absence of a skeleton to scan
from, only minimal scansion will be possible. Rules involving minimal
scansion can see only the immediately adjacent value of [cont] in an
ordered representation of the affricate. This yields the incorrect result
that the cooccurence restrictions depend on the order of the consonants:
cVs would be ruled out, but sVc would be allowed, although in fact both
roots are impossible.
A reviewer has suggested that a claim in Archangeli and Pulleyblank
(1986) would allow them to analyze these constraints while maintaining
412 LINDA LOMBARDI
an o r d e r e d r e p r e s e n t a t i o n o f t h e affricate. A & P suggest t h a t t h e f e a t u r e

h i e r a r c h y d o e s n o t exist in u n d e r l y i n g r e p r e s e n t a t i o n , b u t t h a t s e g m e n t s
are made up of unordered SPE-type feature matrices. They propose that
the feature hierarchy should be eliminated from underlying representation
b e c a u s e it is p r e d i c t a b l e . It is n o t o b v i o u s t h a t t h e i n t e r n a l s t r u c t u r e o f
t h e f e a t u r e s o f a s e g m e n t is t h e t y p e o f r e d u n d a n c y t h a t we s h o u l d e l i m i n -
a t e f r o m u n d e r l y i n g r e p r e s e n t a t i o n . B u t m o r e i m p o r t a n t , this t h e o r y p r e -
dicts t h a t t h e r e s h o u l d b e n o a u t o s e g m e n t a l p r o c e s s e s t h a t h o l d at t h e
level o f u n d e r l y i n g r e p r e s e n t a t i o n . I f u n d e r l y i n g r e p r e s e n t a t i o n consists
o f an u n s t r u c t u r e d list o f f e a t u r e s , t h e n e a c h f e a t u r e d o e s n o t o c c u p y its
o w n a u t o s e g m e n t a l tier, a n d all effects o n u n d e r l y i n g r e p r e s e n t a t i o n
s h o u l d b e effects t h a t h o l d o n l y o f strictly a d j a c e n t s e g m e n t s .
Morpheme structure constraints are clearly autosegmental (Mester
(1986), M c C a r t h y (1985), a n d t h e p r e s e n t analysis o f Y u c a t e c ) , affecting
f e a t u r e v a l u e s t h a t a r e a d j a c e n t on t h e s a m e t i e r while t h e s e g m e n t s
to w h i c h t h o s e f e a t u r e s b e l o n g n e e d n o t t h e m s e l v e s b e a d j a c e n t . Since
m o r p h e m e s t r u c t u r e c o n s t r a i n t s h o l d o f u n d e r l y i n g r e p r e s e n t a t i o n s , this
is a m a j o r p r o b l e m for this a s p e c t o f t h e A r c h a n g e l i a n d P u l l e y b l a n k
t h e o r y . It is n o t p o s s i b l e for t h e t h e o r y to c a p t u r e b o t h t h e l a c k o f o r d e r i n g
a n d t h e a u t o s e g m e n t a l n a t u r e o f m o r p h e m e s t r u c t u r e constraint.16
16 One might imagine a proposal within this theory that MSCs are rules that must apply
immediately at the start of every derivation. However, the theory would also have to say,
then, that the feature hierarchy is not articulated until after this set of rules, since the
unstructured segment representation is needed to capture the fact that these rules operate
on a representation where the values of [cont] are unordered; so an autosegmental analysis
of the constraint would still be impossible. The theory makes it impossible to have effects
of lack of ordering and autosegmental effects simultaneously, because lack of ordering is
achieved as a direct result of lack of hierarchy. In addition, the reviewer claims that since
this theory achieves the effect of underlying lack of ordering, there is no need for [+cont]
and [-cont] to be two privative features - that they can be two values of a single feature
and will be unordered underlyingly due to the lack of hierarchy. However, consideration of
this suggestion points out a difficulty in A&P's theory. They assume that only one value of
a binary feature is specified in underlying representation (a complement rule inserts the
opposite value on all segments unspecified for that feature at the earliest point in the
derivation where that feature is referred to in a rule). Combined with the assumption that
there is no feature hierarchy in UR, this makes it impossible for A&P to represent affricates
underlyingly. Given their proposal elsewhere (Archangeli and Pulleyblank (1987)) that affri-
cates have two Supralaryngeal nodes, if the feature hierarchy were underlying, they could
conceivably represent affricates as one Supralaryngeal node marked [+cont] and the other
node unspecified for a value of [cont]. But if underlying representation is only an unstructured
list of features, there is no way in their theory to distinguish an underlying affricate from an
underlying fricative (marked [+cont]) or an underlying stop (unmarked for [cont]) because
both [+cont] and [-cont] cannot be specified in UR.
5.2. Affricates as Segments with Two Supralaryngeal Nodes

Hualde (1988a) argued, as I do here, that affricates are unordered, not
contour segments. However, his proposal differs in a number of important
ways. One major problem, shared by the analysis of Archangeli and
Pulleyblank discussed above, is the proposal that affricates have two
Supralaryngeal nodes, as in (45):
Root
(45) S u p r a l a r y n ~
[-cont] [+cont]
Place
The purpose of the two supralaryngeal nodes in this representation is to
allow the Basque facts presented in section 3.1 to be analyzed as the
delinking of one of the Supralaryngeal nodes. I have shown that this
process can be analyzed without recourse to this. There are a number
of independent arguments against this representation. One of the main
problems with it is that it does not explain why the two parts of the
affricate always have the same Place features. In the representation of the
affricate proposed in this paper and the representation of Sagey (1986),
there is only one Place node, and so the two 'parts' of the affricate will
necessarily have the same Place features. But in a representation with two
Supralaryngeal nodes, each will have its own Place node, and there is no
obvious reason why they could not be different. This issue is not addressed
in Hualde (1988a). In an earlier paper that assumes an ordered representa-
tion of affricates, Hualde (1987) claims that this is an advantage of his
analysis, in that there are affricates in which the two parts of the affricate
differ in Place, for example [~]. However, both parts of this segment are
[labial], with the difference between labial and labiodental likely being
merely a property of the phonetic implementation of the segment in a
particular language. Segments such as [b~] are also mentioned. However,
as Sagey (1986) demonstrates, these are complex segments, and should
only be represented with a degree of closure feature for their major
articulator; they should not be represented with two values of [cont].
Another problem with the representation in (45) is the fact that evidence
for the existence of the Supralaryngeal node is not at all convincing. As
McCarthy (1988) points out, for the Laryngeal node and Place node, there
is ample evidence from the effects of spreading and delinking, and OCP
414 LINDA LOMBARDI
effects, that an abstract node is required. But there is no clear evidence

that these processes involve the Supralaryngeal node. The one process
that is supposed to result from delinking of the Supralaryngeal node is
reduction of a glottalized consonant such as [t'] to glottal stop. However,
this can be analyzed as delinking of the Place node, a process that the
geometry predicts and that would otherwise not exist. Delinking of the
Place node would entail loss of Manner distinctions. As I discuss above,
loss of Manner features entails loss of Place features, because without
Manner, it is impossible to realize Place. McCarthy makes the complemen-
tary observation that loss of Place features entails loss of Manner features
because the segments that do not have Place features, [h] and [~], cannot
support a distinction in Manner. With this process analyzed as delinking
of Place, and considering evidence that [nasal] must be dominated by the
Root node, McCarthy concludes that there is no evidence for the existence
of a Supralaryngeal node.
Hualde (1988b) uses another representation of affricates. In this work,
affricates are represented as consisting of two Root nodes; the rest of
the analyses of Hualde (1988a) are essentially preserved, including the
parameter set for each rule. It is mentioned that a representation like that
of Clements and Keyser (1983) is unsatisfactory because it does not ac-
count for the fact that the two 'parts' of the affricate share all features
except the values of [cont], but the implications of this for the representa-
tion with two Root nodes are not addressed. In fact, Hualde's (1988b)
representation is essentially identical to that of Clements and Keyser
(1983), and the same objections will hold of it (see section 5.0 above).
5.2.1. Basque Stop Voicing and Palatalization
Hualde (1988a, b) concludes on the basis of certain processes in Basque,

that "we must stipulate the existence of a parameter whose setting will
determine whether a rule taking segments characterized as [o~cont] will
also apply to [cecont]/[-~cont] segments". This section examines the evi-
dence for this parameter, which I will refer to as the Exclusivity parameter.
There is a rule of stop voicing in Basque that takes place after a nasal
or lateral. Hualde (1988a) states the rule as in (46):
(46) Basque Stop Voicing
[-cont]---~ [+voice]/[-cont, + s o n ] _ _
However, this rule does not apply to affricates, although affricates are
[-cont].
(47) a. Verbal inflection

perfective
afal-tu [afaldu] 'have dinner'
lan-tu [landu] 'labor'
neka-tu [nekatu] 'get tired'
b. Future
neka-tu-ko [nekatuko] 'get tired'
egin-ko [e~iIjgo] 'do, make'
egan-ko [egal3go] 'say'
il-ko [il~o] 'die, kill'
C. Imperfective
neka-tsen [nekatsen] 'get tired'
afal-tsen [afaltsen] 'have dinner'
lan-tsen [lantsen] 'labor'
An obvious explanation for the failure of rule (46) to apply to affricates

is that the rule is blocked by structure preservation, since Basque has no
voiced affricates. Hualde claims that this explanation is not possible,
because there are dialects of Basque that have the voiced affricate, but
do not voice the affricate in this context. He gives three examples of
words containing the affricate, two of which, (b) and (c), appear to contain
the same morpheme:
(48)a. [dzanga] immersion

b. [dzirdzartak] beams
c. [dzartako] blow
An extensive study of the phonology of Basque dialects by Moutard

(1975), which presents the phonemic systems of the dialects states that
there is a Biscayan dialect that has [dz] "~ l'initiale de certaines onomato-
p6es", though Moutard does not list the voiced affricate as a phoneme in
Biscayan. The occurrence of this sound in this type of word does not
invalidate the structure-preservation explanation for the blockage of the
voicing rule with affricates. Sound-symbolic words frequently violate con-
straints that hold true of the rest of phonology of a language, and may
include sounds that are not part of the normal phoneme system of the
416 LINDA LOMBARDI
language; 17 if these sounds occur only in sound-symbolism, they will not

be created by the phonological rules of the language.
Hualde (1988a) also discusses a rule of palatalization in Basque. This
is stated as in (49):
(49) [-cont, +cor] ~ [+high]/[+high, - b a c k ] _ _
The rule is stated as applying to [-cont], so it does not apply to fricatives
(50b). 18 But it does not apply to affricates, which are [-cont] (50c); it
applies only to/t,l,n/(t'=palatalized [t])(50a):
(50)a. Basque
a. itaun [it'aun] 'question'
irudi-tu [irudit'u] 'imagine' (perf)
mutil-a [mutiXe] 'girl'
ipin-i [ipifii] 'put' (perf)
b. gison [gison] 'man'

isen [isen] 'name'
c. itsul-i [itsuli] 'fall' (perf)

alai-tsu [alaytsu] 'happy'
These examples are from the Biscayan dialect of Markina, and Hualde
(1988b) gives additional examples from the nearby dialect of Ondarroa,
where the process is basically the same. These dialects do have the palatal
affricate, so that the application of the rule to affricates cannot be ruled
out by structure preservation. Thus this rule is presented as evidence for
the Exclusivity parameter: the rule has the parameter set to indicate that
it will apply only to segments which are [-cont], excluding segments which
are both [-cont] and [+cont].
Once the rule of Stop Voicing has been reanalyzed, this rule of Palataliz-
ation is the only evidence left for the Exclusivity parameter; no other
rules in the languages I have surveyed reveal the necessity for the marked
17 Samarin (1971) points out that ideophones often have phonemes that occur in no other
parts of speech in the language, and discusses this and other properties of ideophones in
Bantu languages. In Takelma myths (Hymes 1979),/~z/is prefixed to some words spoken by
the bear; this is a sound which is not otherwise found in Takelma. English onomatopoeic
oink includes a complex peak followed by a complex coda containing a noncoronal consonant;
this syllable structure is not otherwise possible in English (Selkirk (1982)). Sound-symbolic
words can also have restrictions on their form which do not hold of the rest of the language;
see Bloomfield (1933) on the morphology of sound-symbolic roots in English.
18 See Hualde (1988b) for arguments that/1/is [-cont] in Basque, and that/d/is underlyingly
unspecified for [cont]. The value of [cont] for voiced stops depends on context, as in Spanish.
'exclusive' setting of the parameter. A parameter that must be set for all
rules referring to affricates in all languages, and yet has the same setting
for all the cases surveyed except this one rule in Basque, seems highly
unlikely. This leads one to suspect that this is not the correct analysis of
the Basque facts; in fact, it is possible to analyze the data without recourse
to any special theoretical device, as I now show.
The phoneme system of Basque differs slightly from one dialect to
another, but in general there are at least four coronal places of articu-
lation. As is shown by the spectrograms and descriptions in Navarro-
Tom~s (1925), which are consistent with descriptions of other dialects in
Moutard (1975) and Hualde (1988b),/t/and/d/are dental;/g,tg/are apico-
alveolar; !s,ts/are laminoalveolar;/g,tg/are palatal or alveopalatal. Thus,
/t,d,g,tg/are apical and/s,ts,g,tg/are laminal. The feature system for these
consonants must be as in (51): 19
(51) t tg,g ts,s tg,g

[distr] + +
[ant] + + + -
The phonemes/1/and/rg are also apicoalveolar ([-distr, +ant]).

In Biscayan dialects, the distinction between/tg/and/ts/has been lost,
and only/ts/remains. Some of these dialects also lack t h e / g / - / s ! distinc-
tion, having only N/. (Hualde 1988a p.18; Moutard part III, p. 33). Since
these dialects lack the apical affricate, the set of apical [-cont] segments
is/t,l,n/; this is exactly the set to which the Palatalization rule applies.
Thus the rule in (49) need only be changed slightly to eliminate the need
for the Exclusivity parameter: if the rule applies to [-cont, cor, -dist]
segments, it will apply only to !t,l,n/in these dialects. Since these dialects
have no affricate with this feature composition, the fact that the rule does
not apply to affricates is unremarkable.
Additional evidence for this analysis comes from the behaviour of Pala-
talization in different dialects. Among the differences in application of
this rule, some dialects palatalize only In,l/; some dialects palatalize/n,1/
and optionally palatalize/t/. This variation in application suggests another
possibility, that the palatalization of the sonorants and o f / t / a r e actually
separate processes. All of the dialects described palatalize/n,1/obligator-
ily; this would be a rule applying to [cor, -cont, son] ([-cont] prevents
the rule from applying to/r/). Another rule palatalizes the dental stops;
19 Hualde lists /s/ as [ - a n t ] and /g/ as [+ant], but this is inconsistent with the definition of
the feature [anterior].
418 LINDA LOMBARDI
this rule occurs in some dialects and not others, and may be optional.
Because none of the affricates in any dialect are dental, the rule would
not be expected to apply to them. However, it is possible that the optional
rule which palatalizes obstruents does not have to be restricted to dentals,
but can apply to all apicals, as does the rule of sonorant palatalization.
This is suggested by the data in Iverson and Ofiederra (1985). The dialect
of San Sebastian is one in which palatalization of/n,1/is obligatory, and
that o f / t / is optional. Iverson and Ofiedarra point out several relevant
facts:
1. The rule applies only t o / t / and no other apical obstruent because
this is the only apical obstruent that happens to appear in the relevant
environment. In Basque, stems ending in/i/or/y/and a coronal consonant,
that consonant is/n,1/and, more infrequently,/t/, but not, for example,
/~,d/. And the initial consonant of suffixes, when alveolar, is only/l,n,t/.
2. Laminal sibilants do occur in the correct environment, but they do
not undergo palatalization.
3. One suffix begins with the apical affricate/t~/, and this suffix also
undergoes Palatalization; the rule is optional and 'stylistically equivalent'
to palatalization of/t/(footnote 5, p. 60).
Thus it would seem that the rule which optionally palatalizes obstruents
in this dialect applies to all apical obstruents. The fact that it is not seen
to apply to/~/is due to a lexical gap. The rule does not apply to laminals,
which suggests that an analysis restricting it to laminals in other dialects
is not unreasonable; in those dialects which do not have the apical affri-
cate, then, the rule will not apply to affricates. In dialects which do have
the apical affricate, like the one of San Sebastian, it behaves exactly like
/t/ with respect to this rule. No special theoretical device is needed to
account for the behavior of affricates; in fact, such a device would be
empirically as well as theoretically undesirable.
5.3. Affricates as Right-Headed Contour Segments

Shaw (1987) proposes, assuming an ordered representation of the affri-
cate, that the right branch (the [+cont] branch) acts as the head of the
segment. As evidence for this, the following data from Nisgha, a language
of British Columbia, is presented. Nisgha has a CVC reduplicative prefix,
exemplified in (52):
(52) Nisgha
t'a:p t°ip-t'a:p 'to drive something in'
qa:p qap-qa:p 'piece'
9ux ~ax-gux 'to throw something'
The following examples are given, in which the Root has an affricate in
the final C position. In these cases, the affricate appears as a fricative in
the reduplicated prefix.
(53) Nisgha
pats pis-pats 'to lift something'
k'ats-k w k'is-kats-kw 'to have arrived (boat, vehicle)'
q'uts q'as-quts 'to cut something'
9iits 9as-~iits 'to fry, iron something'
hits has-hits 'to send something to someone'
Under Shaw's theory, in mapping the Root melody to the reduplicative
prefix, the first half of the affricate is 'skipped over' and only the second
half is copied. This is interpreted as evidence for the affricate being right-
headed.
However, no evidence is given that this is an effect of mapping and not
just a phonological rule. An obvious alternative is that this is simply a
rule of spirantization. (The rule may be lexically conditioned; no data
bearing on this question are presented.)
There is a rule of Dorsal Spirantization that occurs in the same position:
(54) Dorsal Spirantization

k,kW,q~ x,xW,X/ C
t'ak t'ix-t'ak 'to forget something'
lukw luxW-lukw 'to move something'
tsooq tsaX-tsooq 'to be embarrassed'
Glottalized consonants deglottalize in the same position; this is another

process that is analyzed as an effect of the process of mapping to the
template, but that could as well be a phonological rule of deglottalization.
(55) hit' hat-hit' 'to stick'

tam' tim-tam' 'to press something'
Glottalized affricates undergo both rules, both spirantization and deglot-

talization.
(56) hats' has-hats' 'to bite something'

The analysis involving right-headed affricates assumes that in both Deaffri-
cation and Deglottalization, the head branch, which is the right branch,
is what is copied in the second C position of the prefix. Assuming an
ordered affricate, it makes sense to refer to a right branch. But to extend
420 LINDA LOMBARDI
this notion of right-headedness to glottalized consonants, it is necessary

to call the Supralaryngeal branch the "right" branch and the Laryngeal
branch the "left" branch in the glottalized consonants. Since glottalization
and the Supralaryngeal features are not ordered with respect to each other
in any other sense, even the superficial justification that exists for ordering
the parts of the affricate does not exist in the case of glottalized Cs.
Since affricates are copied intact into initial position in the reduplicative
affix, certain language-specific stipulations about the nature of the final
consonant position of the affix must still be made: that certain positions
are specified to copy only the "head" of a segment. This does not seem
to have any clear advantage over postulating a phonological rule affecting
those positions.
Note that it is typical, cross-linguistically, for affricates that undergo
spirantization processes to turn to fricatives. An example is the process
traditionally called the Gorgia Toscana that occurs in some Italian dialects
(Izzo 1972). Intervocalically, stops become [h] and affricates become the
homorganic fricative:
(57) Italian Tuscan dialect
amiko amiho 'friend'
fwoko fwoho 'fire'
pa~e page 'peace'
no6i nogi 'nuts'
Neutralization of laryngeal features syllable- or word-finally is quite com-
mon cross-linguistically; in addition, Nisgha has rules of deglottalization
in other positions as well as the second consonant position of this prefix
(see Tarpent 1983). Thus, proposing a rule of deglottalization to account
for these facts about reduplication does not seem at all implausible.
It would seem that the correct approach to take to the Nisgha data
would be to try to achieve a generalization over the rules of spirantization
that occur in this preconsonantal, final position of the reduplicative tem-
plate. An approach that uses the notion of "head" and the properties of
the mapping procedure requires otherwise unsupported stipulations about
the ordering of the Supralaryngeal and Laryngeal nodes; also, no support
is given for the assumption that these effects are actually due to mapping
principles and are not simply phonological rules.
6. C O N C L U S I O N
A close examination of the behavior of affricates shows that previous
analyses, which included the surface ordering of the affricate as part of
underlying representation, must be incorrect. In fact, not only are the

values of [cont] in the affricate unordered, but they are not two values of
the same feature; rather, there must be two single-valued [stop] and
]cont] features. Under this proposal, it is possible to handle without any
additional stipulations all of the cases that have been discussed in recent
literature concerning the representation of affricates. In addition, it is
possible to analyze cases which are a problem for earlier theories, such
as the morpheme structure constraints of different varieties of Yucatec
Maya. This proposal also allows new insight into a phenomenon not
previously considered relevant to this question, that of nasal deletion
before fricatives in various languages. The central arguments can be briefly
restated as follows:
1. The ordering of the stop and fricative parts of the affricate is never
contrastive; no language contains single segments of the form [+cont]
[-cont]. Under the assumption that lexical entries should not contain
predictable information, this ordering should not be part of the underlying
representation.
2. Phonological processes involving affricates do not show edge effects.
Rules that apply to segments that are [-cont] apply to affricates, as do
rules applying to [+cont] segments, regardless of the edge that the rule
context is on.
3. There are morpheme structure constraints that show that the values
of [cont] are not only unordered but must be on separate tiers. If they
are on separate tiers, the two 'values' must be separate single-valued fea-
tures; since features on different tiers are not ordered with respect to one
another in a single segment, the lack of ordering that is needed to explain
the phonological behavior of affricates will follow.
4. Those processes that show authentic edge effects can be demon-
strated to be phonetic. This is expected, since the affricate is ordered at
the surface.
422
APPENDIX: ROOT MORPHEMES OF CLASSICAL YUCATEC
Second C of C V C root:
First C of C V C root: p t c k p' t' c' 6' k' s g
IV
p 5 6 2 6 6 0 0 4 3 3 5 4
IV
t 5 5 1 5 7 5 0 5 2 2 4 4
II I IV I II I
c 2 0 4 0 4 0 0 0 0 2 0 0
II I I IV I II
1 0 0 4 6 1 0 0 0 5 0 0
IV
k 5 6 6 6 7 2 0 1 2 0 4 4
IV III III III III

p' 1 4 0 2 6 1 0 0 0 0 6 3
IV III III III III

t' 1 0 1 3 2 0 5 0 0 0 1 1
IV I III III III III I

c' 2 3 0 0 5 0 0 3 0 0 2 0
I IV III III III III I

6' 3 2 0 1 6 0 0 0 5 0 0 2
IV III III III III

k' 6 4 1 7 1 0 0 0 0 4 6 6
I I I
s 5 6 0 0 5 4 1 6 0 3 10 0
I I I
4 3 0 2 7 3 4 0 3 5 0 2
Key to table:
Each box contains a n u m b e r that is the n u m b e r of roots of that form. Some boxes also
contain a r o m a n n u m e r a l above that n u m b e r . These numerals correspond to the m o r p h e m e
structure constraints; unless otherwise indicated below, they indicate the constraint that rules
out that form, as follows:
I. Anterior harmony.
II. Constraint on ordering of affricates.
III. If both Cs are glottalized, they must be identical.
IV. If there are two Cs in a root that are identical except for glottalization, the glottalized
C m u s t be first. In the table, both orders - both occuring and nonoccurring - are marked.
LINDA LOMBARDI 423
REFERENCES
Anderson, S.: 1974, The Organization of Phonology, Academic Press, New York.
Archangeli, D. and D. Pulleyblank: 1986 The Content and Structure of Phonological Repre-
sentations, unpublished manuscript, University of Arizona and University of Southern
California.
--: 1987, 'Maximal and Minimal Rules: Effects of Tier Scansion', Proceedings of NELS
17, 16-35.
Barlow, A. R.: 1951, Studies in Kikuyu Grammar and Idiom, William Blackwood and Sons,
Edinburgh.
Bloomfield, L.: 1933, Language, University of Chicago Press, Chicago.
Borowsky, T: 1987, 'Antigemination in English Phonology', Linguistic Inquiry 18, 671-78.
Browman, C. and L. Goldstein, 1986, 'Towards an Articulatory Phonology', Phonology
Yearbook 3, 219-52.
Chomsky, N. and M. Halle: 1968, The Sound Pattern of English. Harper and Row, New
York.
Clements, G. N.: 1985, 'The Geometry of Phonological Features', Phonology Yearbook 2,
225-52
Clements, G. N. and S. J. Keyser: 1983, CV Phonology: A Generative Theory of the Syllable,
MIT Press, Cambridge, Massachusetts.
Diccionario de Motul, Maya-Espafiol, atribuido a Fray Antonio de Ciudad Real, 1929, edited
by J. M. Hernandez. Talleres de la Compania Tipografica Yucateca, Mdrida, Yucatan,
Mexico.
Diccionario Maya Cordemex, 1980, Ediciones Cordemex, Mdrida, Yucatan, Mexico.
Elson, B.: 1947, 'Sierra Popoluca Syllable Structure', IJAL 13, 13-17.
Foster, M. L. and G; Foster: 1948, Sierra Popoluca Speech, Smithsonian Institution Institute
of Social Anthropology Publication No. 8, United States Government Printing Office,
Washington, DC.
Goldsmith, J.: 1976, Autosegmental Phonology, unpublished Ph.D dissertation, MIT, Cam-
bridge, Massachusetts.
Hayes, B.: 1986, 'Inalterability in CV Phonology', Language 62, 321-51.
:1988, 'Dipthongization and Coindexing', unpublished manuscript, UCLA.
Hoard, J.: 1978, 'Syllabification in Northwest Indian Languages, With Remarks on the
Nature of Syllabic Stops and Affricates', in A. Bell and J. B. Hooper (eds.), Syllables and
Segments, North-Holland, Amsterdam, pp. 59-72.
Hualde, J.: 1987, 'On Basque Affricates', Proceedings of WCCFL VI, 77-89.
: 1988a, 'Affricates are not Contour Segments', Proceedings of WCCFL VII, 77-89.
: 1988b, A Lexical Phonology of Basque, unpublished Ph.D. dissertation, UCLA.
Hymes, D.: 1979, 'How to Talk Like a Bear in Takelma', 1JAL 45, 101-106.
It6, J.: 1986, Syllable Theory in Prosodic Phonology, Ph.D. dissertation, distributed by
Graduate Linguistics Student Association, University of Massachusetts, Amherst.
Iverson, G. K. and M. L. Ofiederra: 1985, 'On Basque Palatalization', Folia Linguistica
16, 51-61.
Izzo, H.: 1972, Tuscan and Etruscan, University of Toronto Press, Toronto.
Jakobson, R., G. Fant, and M. Halle: 1951, Preliminaries to Speech Analysis', MIT press,
Cambridge, Massachusetts.
Kaye, J.: 1985, 'On the Syllable Structure of Certain West African Languages', in D. L.
Goyvaerts (ed.), African Linguistics: Essays in Memory of M. W. K. Semikenke, John
Benjamins, Amsterdam, pp. 285-308.
Keating, P.: 1988, 'A Survey of Phonological Features', Distributed by Indiana University
Linguistics Club, Bloomington.
Kenstowicz, M.: 1972, 'Lithuanian Phonology', Studies in the Linguistic Sciences (University
of Illinois-Urbana) 2, 1-85.
424 THE NONLINEAR ORGANIZATION OF THE A F F R I C A T E
Kiparsky, P.: 1982, 'Lexical Morphology and Phonology', in Linguistic Society of Korea
(ed.), Linguistics in the Morning Calm, Hanshin Publishing, Seoul, pp. 3-91.
Ladefoged, P.: 1968, A Phonetic Study of West African Languages, Cambridge University
Press, Cambridge.
- - : 1971, Preliminaries to Linguistic Phonetics, University of Chicago Press, Chicago.
Lass, R.: 1976, English Phonology and Phonological Theory, Cambridge University Press,
Cambridge.
Leben, W.: 1973, Suprasegmental Phonology, unpublished Ph.D. dissertation, MIT, Cam-
' bridge, Massachusetts.
Levin, J.: 1987, 'Between Epenthetic and Excrescent Vowels', M. Crowhurst (ed.), Proceed-
ings of WCCFL 6, Stanford Linguistic Association, pp. 187-201.
Lewis, G. L.: 1967, Turkish Grammar, Oxford University Press.
McCarthy, J. J.: 1985, 'Features and Tiers: The Structure of Semitic Roots', Talk presented
at MIT, Cambridge, Massachusetts.
- - : 1986, 'OCP Effects: Gemination and Antigemination', Linguistic Inquiry 17, 207-63.
- - : 1988, 'Feature Geometry and Dependency: A Review,', Phonetica 43, 84-108.
- - : 1989, 'Linear Ordering in Phonological Representation', Linguistic Inquiry 20, 71-99.
McCarthy, J. J. and A. S. Prince: 1990, 'Foot and Word in Prosodic Morphology: The
Arabic Broken Plural', NLLT 8, 2.
McQuown, N.: 1967, 'Classical Yucatec', in Robert A. Wauchope (ed.), Handbook of
Middle American Indians, Vol 5; and Norman McQuown (ed.), Linguistics, University of
Texas Press, Austin, pp. 201-47.
Mascar6, J.: 1984, 'Continuant Spreading in Basque, Catalan and Spanish', in M. Aronoff,
and R. T. Oehrle (eds.), Language Sound Structure, MIT Press, Cambridge, Massachu-
setts, pp. 287-298.
Mester, R. A.: 1986, Studies in Tier Structure, Ph.D dissertation, distributed by Graduate
Linguistics Student Association, University of Massachusetts, Amherst.
Mohrlang, R.: 1972, Studies in Nigerian Languages No. 2: Higi Phonology, Institute of
Linguistics, Zaira.
Moutard, N.: 1975, 'Etude phonologique sur les dialectes basques', Fontes Linguae Va-
sconum 19, 5-42; 2, 141-90; 22, 9-54.
Navarro Tom,is, T.: 1925, 'Pronunciacion Guipuzcoana', in Homena]e a Menendez Pidal,
Madrid.
Nemeth, J.: 1962, Turkish Grammar, Mouton, The Hague.
Prince, A.: 1987, 'Planes and Copying', Linguistic Inquiry 18, 491-509.
Rosenthall, S .: 1988, 'The Representation of Prenasalized Consonants', Proceedings of
WCCFL VII, 277-291.
Rubach, J.: 1986, 'Does the Obligatory Contour Principle Operate in Polish?', Studies in
the Linguistic Sciences 16, 133-47
Sagey, E.: 1986, The Representation of Features and Relations in Nonlinear Phonology,
unpublished Ph.D. dissertation, MIT, Cambridge, Massachusetts.
Samarin, W.: 1971, 'Survey of Bantu Ideophones', African Language Studies 12, 130-68.
Selkirk, L.: 1982, 'The Syllable', in van der Hulst and Smith (eds.), The Structure of
Phonological Representations, Foris, Dordrecht.
: 1988, 'A Two-Root Theory of Geminates', talk presented at NELS 19.
Shaw, P.: 1987, 'Non-conservation of Melodic Structure in Reduplication', CLS 23, 291-306.
Steriade, D.: 1987, 'Locality Conditions and Feature Geometry', Proceedings of NELS 17,
595-617. Graduate Linguistics Student Association, Amherst, Massachusetts.
: forthcoming, 'On Class Nodes', to appear in B. Hayes (ed.), Segment Structure,
Academic Press, New York, (cited in Hayes 1988).
Straight, H.: 1976, The Acquisition of Maya Phonology: Variation in Yucatec Child Language,
Garland, New York.
425
Swadesh M., M. C. Alvarez, and J. R. Bastarrachea: 1970, Diccionario de elementos de

Maya Yucateco colonial, Universidad Nacional Autdnoma de M6xico.
Tarpent, M.-L.: 1983, 'Morphophonemics of Nisgha Plural Formation: a Step Toward Proto-
Tsimshian Reconstruction', Kansas Working Papers in Linguistics 8.2 123-214
Wonderly, W.: 1951, 'Zoque: Phonemics and Morphology', IJAL 17, 1-9; 18, 105-23.
Yip, M.: 1989, 'Feature Geometry and Cooccurence Restrictions', Phonology 6, 349-374.
Received 9 December 1988

Revised 26 June 1989
Department of Linguistics
South College
University of Massachusetts
Amherst, MA 01003
U.S.A.

Lombardi_1990_The_nonlinear_organization_of_the_affricate

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lombardi_1990_The_nonlinear_organization_of_the_affricate

Uploaded by

Copyright:

Available Formats

LINDA LOMBARDI

THE NONLINEAR ORGANIZATION OF

Affricates are segments which are composed, on the surface, of ordered

This representation is proposed in order to account for rules that show

Natural Language and Linguistic Theory 8: 375425, 1990.

[cont], the remainder of the feature geometry assumed here is that of

1. FURTHER DETAILS OF THE FEATURE [CONTINUANT]

Current theories of feature geometry (Mester 1986, Sagey 1986, McCarthy

seem to have a special relationship to one another; for example, weaken-

2. PHONOLOGICAL PROCESSES INVOLVING AFFRICATES

Two types of phonological rules involving affricates have been distin-

Given the representation of affricates in (2), rules that operate on

[-cont] Place [-cont]

will result in an ill-formed segment. As explained above in section 1,

3.3. Zoque Voicing and Nasal-Consonant Clusters

This rule is presented by Sagey (1986) as an example of an edge effect:

(12) s~ine 'conscience'

The nasal assimilation rule assimilates [n] to a following [-cont] segment.

are not subject to stray erasure, s As predicted by the condition in (13),

3.4. Modern Yucatec Maya Morpheme Structure Constraints

consonants in a root. One of these is stated in (15):

2. The language has a prohibition against a branching [+cont]. Thus

[+cont] [-cont] [-cont] [+cont]

[-cont] [+cont] [+cont] [+cont] [-cont] [+cont]

3.5. Morpheme Structure Constraints in Classical Yucatec

3.5.1. Constraint on Ordering of Affricates: the Lexical Entries of These

Considering only nonglottalized coronal consonants, the following con-

lexical entries, including redundancies in linear order (McCarthy 1989),

A stop-affricate root such as tVc will also be represented as a single

(23) Root / ~ , , ~ [+cont] (unassociated)

(25)a. Root node spreads to fulfill template:

(25)b. Associate unlinked feature to the right, changing structure min-

(26)a. Root node spreads to fulfill template:

b. Associate unlinked feature to the right:

,, _.,.,"~'~ ~" [-cont]

3.5.3 The Place and Laryngeal Node in These Roots

(27) kVc cVk sVc cVs

3.5.4. Classical Yucatec MSCs and Contour Segment Theory

As shown in the preceding section, an unordered theory of the affricate

[-cont] [+contl [-contl [-cont] [+cont]

result in crossing association lines, or what is, by the contour segment

Thus, it is possible to explain the ordering of stops and affricates in a

[-cont] [+cont] [-cont] [+cont] [+cont]

[+cont] [-cont] [+cont] [-cont]

Thus it is impossible to achieve a unified analysis of this constraint under

there is no need to assume ordering in order to state this rule; it simply

4. EDGE EFFECTS AND APPARENT EDGE EFFECTS

It is useful to consider what true counter evidence to the present proposal

all be careful to distinguish phonological and phonetic processes. Phonetic

Sagey's analysis is that this is a rule applying to a [-cont] at the end of

2. Voiceless stops are aspirated when followed by a nasal, or when

3. If the first member of a nonhomorganic consonant cluster is a nasal,

The aspiration rule is not structure-preserving; aspiration is not distinctive

merging adjacent identical Place nodes, to avoid a violation of the OCP;

Compare this to labialization in Higi, a Nigerian language. Mohrlang

(41)a. Inflectional Vowel Deletion

b. Progressive Voicing Assimilation

(42) roz + Vz *rozz

Deletion is not only blocked when identical segments would result,

, , ;,:Cma;de]' Identity for

[-ant] [ + ant] Secondary

blocked when the plural morpheme is added to an affricate-final stem.

the least sonorous element first. He alludes to extending this idea to

5.1. Maximal and Minimal Scansion with an Ordered Affricate